WO2023097645A1 - 数据获取方法、装置、设备、介质、芯片、产品及程序 - Google Patents

数据获取方法、装置、设备、介质、芯片、产品及程序 Download PDF

Info

Publication number
WO2023097645A1
WO2023097645A1 PCT/CN2021/135300 CN2021135300W WO2023097645A1 WO 2023097645 A1 WO2023097645 A1 WO 2023097645A1 CN 2021135300 W CN2021135300 W CN 2021135300W WO 2023097645 A1 WO2023097645 A1 WO 2023097645A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
generation model
reference signal
model
channel
Prior art date
Application number
PCT/CN2021/135300
Other languages
English (en)
French (fr)
Inventor
肖寒
Original Assignee
Oppo广东移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Oppo广东移动通信有限公司 filed Critical Oppo广东移动通信有限公司
Priority to PCT/CN2021/135300 priority Critical patent/WO2023097645A1/zh
Publication of WO2023097645A1 publication Critical patent/WO2023097645A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines

Definitions

  • the embodiments of the present application relate to the technical field of mobile communication, and specifically relate to a data acquisition method, device, device, medium, chip, product, and program.
  • AI artificial intelligence
  • Embodiments of the present application provide a data acquisition method, device, device, medium, chip, product, and program.
  • An embodiment of the present application provides a data acquisition method, including:
  • the generative model is obtained based on generative confrontation network, real reference signal, and real channel information training;
  • the reference signal sample data and channel information sample data are used to train a channel estimation model.
  • An embodiment of the present application provides a data acquisition device, including:
  • the acquisition unit is configured to acquire reference signal sample data and channel information sample data generated by the generation model; the generation model is obtained based on generation confrontation network, real reference signal, and real channel information training;
  • the reference signal sample data and channel information sample data are used to train a channel estimation model.
  • An embodiment of the present application provides an electronic device, including: a memory and a processor,
  • the memory stores a computer program executable on the processor
  • the data acquisition method is implemented when the processor executes the program.
  • An embodiment of the present application provides a computer storage medium, the computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the data acquisition method.
  • An embodiment of the present application provides a chip, including: a processor, configured to call and run a computer program from a memory, so that a device installed with the chip executes the steps of the data acquisition method.
  • An embodiment of the present application provides a computer program product, the computer program product includes a computer storage medium, the computer storage medium stores a computer program, and the computer program includes instructions that can be executed by at least one processor, when the instructions are executed by The at least one processor implements the steps of the data acquisition method when executed.
  • An embodiment of the present application provides a computer program, the computer program causes a computer to execute the data acquisition method.
  • the embodiment of the present application provides a data acquisition method.
  • the data acquisition device can acquire the reference signal sample data and channel information sample data generated by the generation model; wherein, the generation model is based on the generation confrontation network, the real reference signal, and the real channel The information is trained; in addition, the reference signal sample data and the channel information sample data are used to train the channel estimation model.
  • the reference signal sample data and channel information sample data in this application are generated by a well-trained generative model, and the generative model is constructed based on a generative confrontation network, which requires only a small amount of real reference signal and real channel information. Realize the training of the generative model, avoiding the process of collecting a large number of real reference signals and real channel information and modeling the channel, and greatly reducing the difficulty of data acquisition and manual overhead.
  • FIG. 1 is a schematic diagram of a communication flow of a wireless communication system provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of a channel estimation and recovery process in a wireless communication system provided by an embodiment of the present application
  • Fig. 3 is a schematic structural diagram of a neural network proposed by the related art
  • FIG. 4 is a schematic structural diagram of a convolutional neural network provided by related technologies
  • FIG. 5 is a schematic diagram of implementing channel estimation and restoration by using AI provided by an embodiment of the present application.
  • FIG. 6 is a first schematic flow diagram of a data acquisition method provided by an embodiment of the present application.
  • FIG. 7 is a second schematic flow diagram of a data acquisition method provided by an embodiment of the present application.
  • FIG. 8 is a schematic diagram of an interface between a reference signal generation model and a channel generation model conditioned on the reference signal generation model provided by an embodiment of the present application;
  • FIG. 9 is a schematic diagram of a reference signal data structure provided by an embodiment of the present application.
  • FIG. 10 is a schematic diagram of a channel information data structure provided by an embodiment of the present application.
  • FIG. 11 is a schematic diagram of an interface between a channel generation model and a channel-conditioned reference signal generation model provided by an embodiment of the present application;
  • FIG. 12 is a schematic diagram of a training process of a reference signal generation model provided by an embodiment of the present application.
  • FIG. 13 is a first schematic diagram of a network structure of a generative confrontation network provided by an embodiment of the present application.
  • FIG. 14 is a schematic diagram of a training process of a channel generation model provided in an embodiment of the present application.
  • FIG. 15 is a second schematic diagram of a network structure of a generative confrontation network provided by an embodiment of the present application.
  • FIG. 16 is a schematic diagram of a training process of a channel generation model conditioned on a reference signal provided in an embodiment of the present application
  • FIG. 17 is a schematic diagram of a training process of a channel-conditioned reference signal generation model provided by an embodiment of the present application.
  • Fig. 18 is a schematic structural diagram of a data acquisition device provided by an embodiment of the present application.
  • Fig. 19 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • Fig. 20 is a schematic block diagram of a chip provided by an embodiment of the present application.
  • FIG. 1 is a schematic diagram of a communication flow of a wireless communication system provided by an embodiment of the present application.
  • the wireless communication system may include a transmitting end and a receiving end.
  • the transmitter 101 performs channel coding and modulation on the source bit stream to obtain modulated data; inserts a reference signal into the modulated data, and the inserted reference signal is used for channel estimation at the receiving end, and finally forms a sending signal, which reaches the receiving end through the channel end.
  • the transmission signal will be interfered by noise during the process of transmitting the signal to the receiving end through the channel.
  • the receiver 102 first receives the signal transmitted by the sending end to obtain the received signal, and then uses the reference signal in the received signal to perform channel estimation to obtain channel state information (Channel State Information, CSI).
  • the receiving end feeds back the CSI to the sending end through the feedback link for the transmitter to adjust channel coding, modulation, precoding, etc.
  • the receiver obtains the final restored bit stream by demodulating the received signal and channel decoding. .
  • Figure 1 is a simple illustration of the communication process of the wireless communication system.
  • modules not listed in the wireless communication system, such as resource mapping, precoding, interference cancellation, and CSI measurement. It is designed and implemented separately, and then each independent module can form a complete wireless communication system after integration.
  • the aforementioned wireless communication system may be a Long Term Evolution (LTE) system, LTE Time Division Duplex (Time Division Duplex, TDD), Universal Mobile Telecommunication System (UMTS), Internet of Things (Internet of Things, IoT) system, Narrow Band Internet of Things (NB-IoT) system, enhanced Machine-Type Communications (eMTC) system, 5G communication system (also known as new wireless (New Radio, NR) communication system), or future communication systems (such as 6G communication systems), etc.
  • LTE Long Term Evolution
  • TDD Time Division Duplex
  • UMTS Universal Mobile Telecommunication System
  • IoT Internet of Things
  • NB-IoT Narrow Band Internet of Things
  • eMTC enhanced Machine-Type Communications
  • 5G communication system also known as new wireless (New Radio, NR) communication system
  • future communication systems such as 6G communication systems
  • the receiver's estimation and recovery of the wireless channel will directly affect the data reception, thereby affecting the performance of the communication system.
  • Figure 2 is a schematic diagram of the channel estimation and recovery process in a wireless communication system provided by the embodiment of the present application.
  • the signal sent by the transmitter on the time-frequency resource as shown in (a) except for information data symbols , a series of specific pilot symbols known to the receiver (ie reference signal symbols) are also transmitted.
  • the reference signal may be a channel state information reference signal (Channel State Information-Reference Signal, CSI-RS) signal, a demodulation reference signal (DeModulation Reference Signal, DMRS) signal, etc.
  • CSI-RS Channel State Information-Reference Signal
  • DMRS Demodulation Reference Signal
  • the signal sent by the transmitter is transmitted to the receiver through the channel.
  • the data symbols and reference signal symbols received by the receiver carry noise (that is, the data symbols and reference signal symbols carrying noise), and the receiver can carry noisy data symbols and reference signal symbols are used for channel estimation.
  • the receiver can use the least squares method (Least Squares, LS) or the minimum mean square error (Minimum Mean Square Error, MMSE) and other methods estimate the channel information at the time-frequency position of the reference signal. Then the receiver can perform channel recovery based on the channel information.
  • the receiver uses the interpolation algorithm to recover the channel information on the full time-frequency resource according to the channel information estimated at the symbol position of the reference signal. Channel information, used for subsequent channel information feedback or data recovery, etc.
  • the resource where the channel has been estimated/restored is the resource at the position of the reference signal symbol. It can be seen from (d) that the resources where the channel has been estimated/restored are the resources at the positions of the reference signal symbols and the data symbols.
  • Fig. 3 is a schematic structural diagram of a neural network proposed by related technologies.
  • the structure of the neural network may include: an input layer, a hidden layer and an output layer.
  • the input layer is responsible for receiving data, hiding The layer processes the data, and the final result is generated in the output layer.
  • each node represents a processing unit, which can be regarded as simulating a neuron. Multiple neurons form a layer of neural network, and multiple layers of information transmission and processing construct an overall neural network.
  • neural network deep learning algorithms have been proposed in recent years, more hidden layers have been introduced, and feature learning is performed through layer-by-layer training of neural networks with multiple hidden layers, which greatly improves the learning of neural networks.
  • processing capabilities and are widely used in pattern recognition, signal processing, optimization combination, anomaly detection, etc.
  • CNN Convolutional Neural Networks
  • Figure 4 is a schematic structural diagram of a convolutional neural network provided by related technologies.
  • the structure of the convolutional neural network may include: an input layer, multiple convolutional layers, multiple pooling layers, and a fully connected layer and the output layer.
  • the sharp increase of network parameters is effectively controlled, the number of parameters is limited, and the characteristics of local structures are mined, which improves the robustness of the algorithm.
  • FIG. 5 is a schematic diagram of implementing channel estimation and restoration using AI provided by an embodiment of the present application.
  • an AI-based channel estimation model (or called a channel restoration model) can be constructed through a neural network.
  • the channel is estimated and recovered by using the channel estimation model.
  • the input data of the AI-based channel estimation model 51 is a reference signal
  • the output data is the channel information corresponding to the channel transmitting the reference signal (ie, the channel estimation result).
  • other auxiliary information can be added to the input information of the AI-based channel estimation model 51 to improve the performance of the AI-based channel estimation model 51.
  • these other auxiliary information can be reference signals
  • the reference signal input to the AI-based channel estimation model/channel recovery model 51 includes at least the reference signal in the received signal, that is, the reference signal carrying noise.
  • the AI-based channel estimation model can be trained based on reference signals and corresponding channel information. That is to say, the training data set of the AI-based channel estimation model includes a reference signal and channel information corresponding to the reference signal.
  • the most direct way to obtain the above-mentioned training data set is to collect the actual wireless channel, such as obtaining the received reference signal and the channel information corresponding to the reference signal through a paired signal transmitter and signal receiver, or through a specific
  • the receiver collects a signal from a third-party transmitter (such as a cellular network base station) to obtain a reference signal received by the receiver and channel information corresponding to the reference signal.
  • Another method is to use the simulation platform established by the mathematical model of the channel to generate a large amount of training sample data.
  • the AI-based channel estimation model has great dependence and demand on the training data set (including the reference signal and its corresponding channel information). It can be said that the training data set is the key to determine the performance gain of this type of scheme.
  • the wireless communication frequency band is gradually moving from low frequency to high frequency, and gradually moving towards more complex special environments such as air, space, earth and sea.
  • the expansion of more scenarios such as special applications makes the wireless channel environment that the current wireless communication system needs to face more and more complex. Therefore, in the aforementioned complex wireless channel environment, it is very difficult to receive reference signals and collect channel information.
  • the difficulties here include both technical difficulties and operational difficulties.
  • the mathematical modeling of the above-mentioned complex channels is also facing great challenges.
  • the complexity of frequency bands, environments, and scenarios will lead to the complexity of channel modeling. Non-linear channel characteristics and difficult-to-fit channel propagation characteristics will affect the Traditional mathematical modeling to study channels brings difficulties and challenges.
  • an embodiment of the present application provides a data acquisition method.
  • the data acquisition device can acquire reference signal sample data and channel information sample data generated by the generation model; and real channel information training; in addition, reference signal sample data and channel information sample data are used to train the channel estimation model.
  • the reference signal sample data and channel information sample data in this application are generated by a well-trained generative model, and the generative model is constructed based on a generative confrontation network, which requires only a small amount of real reference signal and real channel information. Realize the training of the generative model, avoiding the process of collecting a large number of real reference signals and real channel information and modeling the channel, and greatly reducing the difficulty of data acquisition and manual overhead.
  • the data acquisition method provided in the embodiment of the present application can be applied to the data acquisition device provided in the embodiment of the present application.
  • the data acquisition device can be integrated in electronic equipment through software or hardware, which can be servers, personal computers, industrial computers, etc.
  • the electronic equipment can also be network equipment or terminal equipment with communication functions. Examples are not limited to this.
  • the network device can be an evolved base station (Evolutional Node B, eNB or eNodeB) in the LTE system, or a next-generation radio access network (Next Generation Radio Access Network, NG RAN) device, or a base station in the NR system (gNB), or the wireless controller in the cloud radio access network (Cloud Radio Access Network, CRAN), or the network device can be a relay station, an access point, a vehicle device, a wearable device, a hub, a switch, a bridge , routers, or network devices in the future evolution of the Public Land Mobile Network (Public Land Mobile Network, PLMN), etc.
  • Evolutional Node B, eNB or eNodeB in the LTE system
  • NG RAN next-generation radio access network
  • gNB base station in the NR system
  • CRAN Cloud Radio Access Network
  • the network device can be a relay station, an access point, a vehicle device, a wearable device, a hub, a switch, a bridge , routers,
  • the terminal device 110 may be any terminal device, including but not limited to a terminal device connected to the network device 120 or other terminal devices by wire or wirelessly.
  • the terminal equipment 110 may refer to an access terminal, a user equipment (User Equipment, UE), a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, a user terminal, a terminal, a wireless communication device, user agent, or user device.
  • UE User Equipment
  • Access terminals can be cellular phones, cordless phones, Session Initiation Protocol (SIP) phones, IoT devices, satellite handheld terminals, Wireless Local Loop (WLL) stations, Personal Digital Assistant , PDA), handheld devices with wireless communication functions, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices, wearable devices, terminal devices in 5G networks or terminal devices in future evolution networks, etc.
  • SIP Session Initiation Protocol
  • IoT devices IoT devices
  • satellite handheld terminals satellite handheld terminals
  • WLL Wireless Local Loop
  • PDA Personal Digital Assistant
  • handheld devices with wireless communication functions computing devices or other processing devices connected to wireless modems
  • vehicle-mounted devices wearable devices
  • terminal devices in 5G networks or terminal devices in future evolution networks etc.
  • Fig. 6 is a first schematic flowchart of a data acquisition method provided by an embodiment of the present application. As shown in Fig. 6, the method includes the following contents.
  • Step 610 Obtain reference signal sample data and channel information sample data generated by the generative model; the generative model is trained based on generative adversarial networks, real reference signals, and real channel information; reference signal sample data and channel information sample data are used for training channel estimation Model.
  • the generation model here refers to a pre-trained model.
  • the generative model can be trained based on generative confrontation network, real reference signal, and real channel information.
  • Generative Adversarial Networks is a deep learning model, one of the most promising methods for unsupervised learning on complex data distributions.
  • the generated confrontation network produces the expected output through the mutual game learning between (at least) two models: the Generative Model and the Discriminative Model.
  • the generative model is used to generate virtual data
  • the discriminant model is used to identify the authenticity of the virtual data generated by the generative model (or to identify whether the virtual data is generated by the generative model or from real data).
  • the discriminator model and the generative model can be alternately trained, so that the data generated by the trained generative model can fool the discriminator model. That is to say, for the virtual data generated by the trained generative model, the identification model cannot identify whether the virtual data is generated by the generative model or real data. It can be seen that the virtual data generated by the trained generative model is comparable to the real data.
  • the generation model in this embodiment of the present application can be trained based on real reference signals and real channel information.
  • the generation model in the embodiment of the present application is specifically used to generate virtual reference signal sample data and virtual channel information sample data.
  • the reference signal sample data and channel information sample data generated by it can be very close to the real reference signal and channel information. Therefore, a large number of sample data of reference signals and sample data of channel information can be generated by generating the model to form a training data set for training the channel estimation model.
  • the real reference signal used to train the above generation model may at least include a reference signal received by a receiver (also referred to as a received reference signal), that is, a reference signal carrying noise.
  • a receiver also referred to as a received reference signal
  • reference signals hereinafter refer to received reference signals.
  • the real reference signal is obtained by receiving the channel corresponding to the real channel information. That is to say, the receiver receives the above-mentioned real reference signal on the channel corresponding to the real channel information.
  • the numbers of real reference signals and real channel information used to generate model training are both preset numbers; the preset number is less than or equal to the first threshold.
  • the first threshold may be 50 or 100, which is not limited in this embodiment of the present application.
  • the number of real reference signals and real channel information used to train the generative model provided in this application is far less than the tens of thousands of real reference signals and real channel information required for training the channel estimation model in the related art. That is to say, the embodiment of the present application may only rely on a small amount of real reference signals and real channel information to build a generative model. In this way, the generative model provided by the embodiment of the present application can generate a large number of reference signal sample data and channel information sample data close to the real ones, which are used for the training of the channel estimation model, avoiding the collection of a large number of real reference signals and real channel information and the The channel modeling process greatly reduces the difficulty of data acquisition and manual overhead.
  • the above multiple real reference signals and real channel information may be real data acquired in various frequency bands, radio frequency environments, and wireless channel scenarios.
  • the generative model trained based on real reference signals and real channel information of different frequency bands, radio frequency environments, and wireless channel scenarios can also generate reference signal sample data and channel information corresponding to various frequency bands, radio frequency environments, and wireless channel scenarios sample. In this way, the diversity of reference signal sample data and channel information sample data is guaranteed.
  • the reference signal sample data generated by the generation model and the channel information sample data also have an association relationship.
  • the reference signal sample data and the channel information sample data may be a one-to-one correspondence between the reference signal sample data and the channel information sample data, that is, one reference signal sample data corresponds to one channel information sample data.
  • the reference signal sample data and the channel information sample data having an associated relationship can constitute a virtual reference signal-channel information data pair, which is used for training the channel estimation model.
  • the data acquisition device can acquire the reference signal sample data and channel information sample data generated by the generation model; wherein, the generation model is based on the generation confrontation network, the real reference signal, and the real The channel information is trained; in addition, the reference signal sample data and the channel information sample data are used to train the channel estimation model.
  • the reference signal sample data and channel information sample data in this application are generated by a well-trained generative model, and the generative model is constructed based on a generative confrontation network, which requires only a small amount of real reference signal and real channel information. Realize the training of the generative model, avoiding the process of collecting a large number of real reference signals and real channel information and modeling the channel, and greatly reducing the difficulty of data acquisition and manual overhead.
  • the generative models may include a reference signal generative model, and a channel generative model.
  • the reference signal generation model is used to generate reference signal sample data
  • the channel generation model is used to generate channel information sample data. That is to say, the generation model provided by the embodiment of the present application can use two independent generation models to generate the reference signal sample data and the channel information sample data with correlation respectively.
  • step 610 obtaining the reference signal sample data and channel information sample data generated by the generation model can be achieved through the following steps:
  • Step 6101. Obtain the first data generated by the first generative model
  • Step 6102 Input the first data into the second generation model, and obtain the second data generated by the second generation model.
  • the first generation model may be a reference signal generation model
  • the second generation model may be a channel generation model
  • the channel generation model may also be referred to as a channel generation model conditioned on a reference signal.
  • Fig. 8 shows a schematic interface diagram of a reference signal generation model and a channel generation model conditioned on the reference signal generation model.
  • the output end of the reference signal generation model may be connected to the input end of the channel generation model conditioned on the reference signal.
  • the data obtaining device can firstly obtain the first reference signal sample data (namely the first data) generated by the reference signal generation model, and then, the data obtaining device can input the first reference signal sample data into the in the channel generation model.
  • the channel generation model conditioned on the reference signal can generate first channel information sample data (that is, second data) that has an association relationship with the first reference signal sample data based on the input first reference signal sample data.
  • the reference signal generation model may have no input data, that is, the data acquisition device may directly use the reference signal generation model to generate the first reference signal sample data.
  • the reference signal generation model can also use at least one of the following data as input data:
  • Noise, random numbers, channel type indication information, true reference signal, and statistical information of the true reference signal are examples of the true reference signal.
  • the noise may be noise data from a real environment, or artificially generated noise data, which is not limited in this embodiment of the present application.
  • the random number may be a random number sequence or a pseudo-random number sequence, which is not limited in this embodiment of the present application.
  • the data format of noise and random numbers used as the input data of the reference signal generation model may be one-dimensional vector, two-dimensional matrix, or higher-dimensional data, which is not limited in this embodiment of the present application.
  • the data format of the noise and the random number may be agreed in advance, or may be consistent with the data format of the generated reference signal sample data, which is not limited in this embodiment of the present application.
  • the first indication information is used to indicate the frequency band where the channel is located
  • the second indication information is used to indicate the radio frequency environment
  • the third indication information is used to indicate the wireless channel scenario.
  • the first indication information may indicate whether the current frequency is high or low.
  • the second indication information may indicate that the current radio frequency environment is indoors, outdoors, densely populated cells, or an open field environment.
  • the third indication information may indicate whether the current wireless channel scenario is an Internet of Things scenario, an industrial scenario, or the like.
  • the real reference signal may be actually collected, or a reference signal obtained through mathematical modeling.
  • the statistical information of the real reference signal includes but not limited to the maximum power spectral density, the average power spectral density and the like of the real reference signal.
  • the reference signal generation model can generate the first reference signal sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios based on the above input data, so as to ensure the diversity of the generated first reference signal sample data and richness.
  • a single first reference signal sample data can be composed of three-dimensional data with a size of M*N*P, where the values of M, N, and P They may be equal or not, which is not limited in this embodiment of the present application.
  • the three-dimensional data of M*N*P can also be synthesized into a one-dimensional vector of size 1*(M*N*P) or (M*N*P)*1.
  • the reference signal sample data The data format is not limited.
  • the three dimensions of the first reference signal sample data shown in FIG. 9 may respectively represent a frequency domain dimension, a time domain dimension, and an antenna pair domain dimension.
  • Each unit of the frequency domain dimension corresponds to one subcarrier
  • each unit of the time domain dimension corresponds to one time domain symbol
  • each unit of the antenna pair dimension corresponds to one antenna pair.
  • the reference signal may be represented by a complex number. Therefore, the first reference signal sample data output by the above-mentioned reference signal generation model can add an additional dimension (that is, the fourth dimension) on the basis of the above-mentioned three-dimensional data, and the fourth dimension can represent the real part of the first reference signal sample data and imaginary part.
  • the first reference signal sample data in the embodiment of the present application is not limited to less than four dimensions, and the first reference signal sample data may also include information of more dimensions.
  • the channel generation model conditioned on the reference signal in the embodiment of the present application may input the first reference signal sample data, and its output may be the channel information corresponding to the first reference signal, that is, the first channel Information sample data.
  • the first channel information sample data may consist of data in three dimensions: frequency domain, time domain, and antenna pair domain.
  • the first channel information sample data may be composed of three-dimensional data with a size of I*J*K.
  • the values of I, J, and K may or may not be equal, which is not limited in this embodiment of the present application.
  • the three-dimensional data of I*J*K can also be synthesized into a one-dimensional vector of size 1*(I*J*K) or (I*J*K)*1.
  • the channel information sample data The data format is not limited.
  • the value of I may be the same as that of N
  • the value of J may be the same as that of M
  • the value of P may be the same as that of K.
  • the channel information may also be presented by complex numbers. Therefore, the first channel information sample data output by the above-mentioned channel generation model conditioned on the reference signal can add an additional dimension (that is, the fourth dimension) on the basis of the above-mentioned three-dimensional data, and the fourth dimension can represent the first channel information sample The real and imaginary parts of the data.
  • the first channel information sample data in the embodiment of the present application is not limited to less than four dimensions, and the first channel information sample data may also include information of more dimensions.
  • the first channel information sample data output by the channel generation model conditioned on the reference signal may be channel feature information obtained by mathematical transformation of the original channel information, such as channel feature vector information obtained by SVD decomposition, It may be single-stream channel eigenvector information, or multi-stream channel eigenvector information, such as 2-stream, 4-stream, or 8-stream channel eigenvector information, which is not limited in this embodiment of the present application.
  • the input data may be the first reference signal sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios. Therefore, the above-mentioned channel generation model conditioned on the reference signal may also generate first channel information sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios.
  • a large number of virtual first reference signal sample data can be generated first through the reference signal generation model, and then use the generated first reference signal sample data to generate each channel through the reference signal-conditioned channel generation model.
  • the first channel information sample data corresponding to the first reference signal sample data constitute a plurality of pairs of reference signal-channel information sample data with correlations for training the channel estimation model.
  • the first generation model may be a channel generation model
  • the second generation model may be a reference signal generation model
  • the reference signal generation model may also be referred to as a channel-conditioned reference signal generation model.
  • Fig. 11 shows a schematic interface diagram of a channel generation model and a channel-conditioned reference signal generation model. Wherein, the output end of the channel generation model may be connected with the input end of the reference signal generation model conditioned on the channel.
  • the data obtaining device can firstly obtain the second channel information sample data (i.e. the first data) generated by the channel generation model, and then, the data obtaining device can input the second channel information sample data into the channel-conditioned reference in the signal generation model.
  • the channel-conditioned reference signal generation model can generate second reference signal sample data (that is, second data) that has an association relationship with the second channel information sample data based on the input second channel information sample data.
  • the channel generation model may have no input data, that is, the data acquisition device may directly use the channel generation model to generate the second channel information sample data.
  • the channel generation model can use at least one of the following data as input data:
  • Noise, random numbers, channel type indication information, real channel information, and statistical information of real channel information are examples of real channel information.
  • noise, the random number, and the channel type indication information are all similar to the descriptions in the foregoing embodiments, and for the sake of brevity, details are not repeated here.
  • real channel information may be actually collected, or channel information obtained through mathematical modeling.
  • Statistical information of real channel information includes but not limited to power spectral density and average power spectral density of real channel information.
  • the channel generation model can generate second channel information sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios based on the above input data, so as to ensure the diversity and accuracy of the generated second channel information sample data. richness.
  • a single second channel information sample data is composed of three-dimensional data in the time domain, frequency domain, and antenna pair domain with a size of I*J*K.
  • the values of I, J, and K may or may not be equal, which is not limited in this embodiment of the present application.
  • the three-dimensional data of I*J*K can also be synthesized into a one-dimensional vector of size 1*(I*J*K) or (I*J*K)*1.
  • the channel information sample data The data format is not limited.
  • the second channel information sample data output by the above-mentioned channel generation model can add an additional dimension (that is, the fourth dimension) on the basis of the above-mentioned three-dimensional data, and the fourth dimension can represent the real part and the imaginary part of the second channel information sample data. department.
  • the second channel information sample data in the embodiment of the present application is not limited to less than four dimensions, and the second channel information sample data may also include information of more dimensions.
  • the channel-conditioned reference signal generation model may input the second channel information sample data, and its output may be the reference signal corresponding to the second channel information sample data, that is, the first Two reference signal sample data.
  • the second reference signal sample data may be composed of three dimensions of frequency domain, time domain, and antenna pair domain with a size of M*N*P data composition.
  • the values of M, N, and P may be equal or unequal, which is not limited in this embodiment of the present application.
  • the three-dimensional data of M*N*P can also be synthesized into a one-dimensional vector of size 1*(M*N*P) or (M*N*P)*1.
  • the reference signal sample data The data format is not limited.
  • the value of N may be the same as that of I
  • the value of M may be the same as that of J
  • the value of K may be the same as that of P.
  • the second reference signal sample data output by the above-mentioned channel-conditioned reference signal generation model can add an additional dimension (that is, the fourth dimension) on the basis of the above-mentioned three-dimensional data, and the fourth dimension can represent the second The real and imaginary parts of the reference signal sample data.
  • the second reference signal sample data in the embodiment of the present application is not limited to less than four dimensions, and the second reference signal sample data may also include information of more dimensions.
  • the input data may be second channel information sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios. Therefore, the above-mentioned channel-conditioned reference signal generation model may also generate second reference signal sample data in one or more frequency bands, radio frequency environments, and wireless channel scenarios.
  • a large amount of virtual second channel information sample data can be generated first through the channel generation model, and then use the generated second channel information sample data to generate each first channel through the channel-conditioned reference signal generation model.
  • the second reference signal sample data corresponding to the two channel information sample data constitute multiple pairs of reference signal-channel information sample data with correlations, which are used for training the channel estimation model.
  • the training process of the first generative model is described in detail below.
  • the training process of the first generation model may include the following steps:
  • Step a1 obtaining the third data output by the first generative model to be trained
  • Step a2 input the third data into the first identification model, and output the first identification result through the first identification model; the first identification model is used to identify the probability that the category corresponding to the third data is the category to which the first real data belongs; the first real The data has an association relationship with the first generation model;
  • Step a3. Based on the first identification result, adjust the model parameters of the first generation model to be trained to obtain the first generation model; the probability that the category corresponding to the data generated by the first generation model is the category to which the first real data belongs is greater than the second threshold .
  • the first generation model is used to generate the required data
  • the first identification model is used to identify the authenticity of the data generated by the first generation model (or to identify whether the virtual data is generated by the generation model or from real data). data).
  • the data generated by the trained first generation model can fool the first identification model, that is, the first identification model cannot distinguish the data generated by the first generation model. From real data or generated by the first generative model.
  • the first real data has an association relationship with the first generation model, if the first generation model is a reference signal generation model, then the first real data is a real reference signal, and if the first generation model is a channel generation model, then the first real The data is real channel information.
  • the first generation model may be the reference signal generation model in FIG. 8
  • the corresponding first discrimination model is the reference signal discrimination model
  • the first real data is the real reference signal.
  • the training process of the reference signal generation model can be referred to as shown in FIG. 12 .
  • a reference signal generation model to be trained and a reference signal identification model to be trained are constructed.
  • the generation model of the reference signal to be trained and the identification model of the reference signal to be trained can be composed of one or more network structures of a fully connected network, a convolutional neural network, a residual network, and a self-attention mechanism network.
  • the reference signal generation model to be trained and the reference signal discrimination model to be trained may consist of multiple fully connected layers. Different fully connected layers correspond to different dimensions to extract different data features.
  • the input of the reference signal generation model in the above embodiment is the same, the reference signal generation model to be trained may not have an independent input, and noise, random number, channel type indication information, real reference signal, and real At least one item of statistical information of the reference signal is used as input data.
  • the input of the reference signal identification model to be trained may be the virtual reference signal (ie, the third data) generated by the parameter signal generation model to be trained, and the real reference signal.
  • the model parameters of the reference signal generation model to be trained can be kept unchanged, and the virtual reference signal (ie, the third data) generated by the reference signal generation model to be trained is identified by the reference signal identification model to be trained to determine
  • the category of the current virtual reference signal is the probability of the category of the real reference signal (ie, the first identification result).
  • the model parameters of the reference signal identification model to be trained are adjusted, so that the reference signal identification model to be trained can distinguish the true and false reference signals as much as possible, and a trained reference signal identification model is obtained.
  • the model parameters of the reference signal identification model are kept unchanged, and the virtual reference signal (ie, the third data) generated by the reference signal generation model to be trained is identified through the reference signal identification model to determine the currently generated virtual reference signal.
  • the category of the signal is the probability of the category to which the real reference signal belongs (that is, the first identification result).
  • the model parameters of the reference signal generation model to be trained can be adjusted so that the reference signal discrimination model cannot distinguish the difference between the virtual reference signal generated by the reference signal generation model to be trained and the real reference signal.
  • the reference signal identification model cannot identify whether the virtual reference signal generated by the reference signal generation model to be trained is generated or real, that is, the reference signal identification model identifies the reference signal to be trained
  • the probability that the category corresponding to the virtual reference signal generated by the generation model is the category to which the real reference signal belongs is greater than the second threshold, that is, the reference signal identification model and the reference signal generation model to be trained reach a stable state, complete the training, and obtain a trained reference signal Generate a model.
  • the data acquisition device can separately extract a trained reference signal generation model for generation of reference signal sample data.
  • the first generation model may be the channel generation model in FIG. 11 , the corresponding first identification model is a channel identification model, and the first real data is real channel information.
  • the training process of the channel generation model can be referred to as shown in FIG. 14 .
  • the channel generation model to be trained and the channel identification model to be trained are constructed.
  • the channel generation model to be trained and the channel identification model to be trained can be composed of one or more network structures of fully connected networks, convolutional neural networks, residual networks, and self-attention mechanism networks.
  • the channel generation model to be trained and the channel identification model to be trained may be composed of multiple fully connected layers. Different fully connected layers correspond to different dimensions to extract different data features.
  • the input of the channel generation model in the above embodiments is the same, the channel generation model to be trained may not have an independent input, and may also include noise, random numbers, channel type indication information, real channel information, and real channel information At least one of the statistical information of , is used as input data.
  • the input of the channel identification model to be trained may be virtual channel information (that is, the third data) generated by the channel generation model to be trained, and real channel information.
  • the model parameters of the channel generation model to be trained can be kept unchanged, and the virtual channel information (ie, the third data) generated by the channel generation model to be trained can be identified by the channel identification model to be trained to determine the current virtual channel.
  • the information category is the probability of the category to which the real channel information belongs (that is, the first identification result).
  • the model parameters of the channel identification model to be trained are adjusted, so that the channel identification model to be trained can distinguish the true and false channel information as much as possible, and a trained reference signal identification model is obtained.
  • the model parameters of the channel identification model are kept unchanged, and the virtual channel information (ie, the third data) generated by the channel generation model to be trained is identified through the channel identification model to determine the category of the currently generated virtual channel information is the probability of the category to which the real channel information belongs (that is, the first identification result). Furthermore, for the purpose of increasing the probability, the model parameters of the channel generation model to be trained can be adjusted so that the channel discrimination model cannot distinguish the difference between the virtual channel information generated by the channel generation model to be trained and the real channel information.
  • the channel identification model cannot identify whether the virtual channel information generated by the channel generation model to be trained is generated or real, that is, the channel identification model identifies the information generated by the channel generation model to be trained.
  • the probability that the category corresponding to the virtual channel information is the category to which the real channel information belongs is greater than the second threshold, that is, the channel identification model and the channel generation model to be trained reach a stable state, the training is completed, and the trained channel generation model is obtained.
  • the data acquisition device can separately extract the trained channel generation model for generation of channel information sample data.
  • the training process of the second generation model may include the following steps:
  • Step b1 input the fourth data into the second generation model to be trained, generate fifth data through the second generation model to be trained, and the fourth data is the data generated by the first generation model;
  • Step b2 input the fifth data into the second identification model, and output the second identification result through the second identification model; the second identification result is used to identify the probability that the category corresponding to the combination data is the category to which the second real data belongs;
  • the combination data includes the first Four data and fifth data;
  • the second real data includes real reference signal and real channel information;
  • Step b3 based on the second identification result, adjust the model parameters of the second generation model to be trained to obtain the second generation model; the data formed by combining the data generated by the second generation model with the fourth data has a corresponding category and The probability of the category to which the second real data belongs is greater than the third threshold.
  • the second generation model is used to generate the required data
  • the second identification model is used to identify the authenticity of the data generated by the second generation model (or to identify whether the virtual data is generated by the generation model or from real data).
  • the data generated by the trained second generation model can fool the first identification model, that is, the second identification model cannot distinguish the data generated by the second generation model. From real data or generated by a second generative model.
  • the second generation model may be a channel generation model conditioned on a reference signal as shown in FIG. 8 .
  • Fig. 16 shows the training process of the channel generative model conditioned on the reference signal.
  • a channel generation model to be trained and a channel identification model to be trained may be constructed first.
  • the channel generation model to be trained and the channel identification model to be trained can be composed of one or more network structures of a fully connected network, a convolutional neural network, a residual network, and a self-attention mechanism network.
  • the channel generation model to be trained and the channel identification model to be trained may be composed of multiple fully connected layers. Different fully connected layers correspond to different dimensions to extract different data features.
  • the input data (that is, the fourth data) of the channel generation model to be trained may be a virtual reference signal.
  • the virtual reference signal may be reference signal sample data generated by a trained reference signal generation model. It can be understood that the virtual reference signal is used as the input data of the channel generation model to be trained in this implementation mode, so that the channel generation model to be trained can generate channel information that is associated with the virtual reference signal, thereby obtaining the corresponding reference signal- Channel information sample data pair.
  • the input of the channel discrimination model to be trained in this implementation may include a virtual reference signal (ie, the fourth data), virtual channel information (ie, the fifth data) generated by the channel generation model to be trained, a real reference signal , and real channel information.
  • the virtual reference signal that is, the fourth data
  • the virtual channel information that is, the fifth data
  • the first reference signal sample data whose size is M*N*P shown in FIG. 9 and the first channel information sample data whose size is I*J*K shown in FIG. 10 can be combined as (M +I)*(N+J)*(P+K) three-dimensional data, and input the (M+I)*(N+J)*(P+K) three-dimensional data into the channel discrimination model to be trained.
  • the real reference signal and the real channel information may be jointly input, that is, the real reference signal and the real channel information are combined into real combined data for input.
  • the model parameters of the channel generation model to be trained can be kept unchanged, and the virtual reference signal (i.e. the fourth data) and the virtual channel information generated by the channel generation model to be trained (i.e. fifth data) to identify the combined data, and determine the probability that the category corresponding to the combined data is the category to which the real combined data belongs (ie the second identification result). Furthermore, for the purpose of reducing the probability, the model parameters of the channel identification model to be trained are adjusted, so that the channel identification model to be trained can distinguish the true and false of the combined data as much as possible, and a trained reference signal identification model is obtained.
  • the model parameters of the channel discrimination model are kept unchanged, and the virtual reference signal (ie, the fourth data) and the virtual channel information (ie, the fifth data) generated by the channel generation model to be trained are formed by the channel discrimination model
  • the combination data is identified, and the probability that the category of the current combination data is the category to which the above-mentioned real combination data belongs is determined (ie, the second identification result).
  • the model parameters of the channel generation model to be trained can be adjusted so that the channel discrimination model cannot distinguish the difference between the combination data and the above-mentioned real combination data.
  • the above steps can be repeated. After multiple update iterations, when the channel identification model cannot identify whether the combined data is generated or real, that is, the probability that the channel identification model identifies the category of the combined data as the category of the real combined data is greater than the first
  • the three thresholds that is, the channel identification model and the channel generation model to be trained reach a steady state
  • the training is completed, and the trained channel generation model conditioned on the reference signal is obtained.
  • the data acquisition device can separately extract the trained channel generation model to obtain the channel generation model conditioned on the reference signal, which is used to generate channel information sample data associated with the reference signal sample data.
  • the second generation model may be the channel-conditioned reference signal generation model shown in FIG. 11 .
  • Fig. 17 shows the training process of the reference signal generation model conditional on the channel.
  • a reference signal generation model to be trained and a reference signal identification model to be trained may be constructed first.
  • the generation model of the reference signal to be trained and the identification model of the reference signal to be trained can be composed of one or more network structures of a fully connected network, a convolutional neural network, a residual network, and a self-attention mechanism network.
  • the channel generation model to be trained and the channel identification model to be trained may be composed of multiple fully connected layers. Different fully connected layers correspond to different dimensions to extract different data features.
  • the input data (that is, the fourth data) of the reference signal generation model to be trained may be virtual channel information.
  • the virtual channel information may be channel information sample data generated by a trained channel generation model. It can be understood that the virtual channel information is used as the input data of the reference signal generation model to be trained in this implementation mode, so that the reference signal generation model to be trained can generate a reference signal associated with the virtual channel information, thereby obtaining the corresponding reference Signal - channel information sample data pair.
  • the input of the reference signal identification model to be trained in this implementation may include virtual reference channel information (that is, the fourth data), the virtual reference signal generated by the reference signal generation model to be trained (that is, the fifth data), Real reference signal, and real channel information.
  • the virtual channel information (that is, the fourth data) and the virtual reference signal (that is, the fifth data) can be jointly input, that is, the virtual channel information and the virtual reference signal are combined into A set of data inputs.
  • the first reference signal sample data whose size is M*N*P shown in FIG. 9 and the first channel information sample data whose size is I*J*K shown in FIG. 10 can be combined as (M +I)*(N+J)*(P+K) three-dimensional data, and input the (M+I)*(N+J)*(P+K) three-dimensional data into the channel discrimination model to be trained.
  • the real reference signal and the real channel information may be jointly input, that is, the real reference signal and the real channel information are combined into real combined data for input.
  • the model parameters of the reference signal generation model to be trained can be kept unchanged, and the virtual channel information (that is, the fourth data) and the virtual reference generated by the reference signal generation model to be trained can be identified by the reference signal to be trained.
  • the combined data composed of the signal ie, the fifth data
  • the probability ie, the second identification result
  • the model parameters of the reference signal identification model to be trained are adjusted, so that the reference signal identification model to be trained can distinguish the true and false of the combined data as much as possible, and a trained reference signal identification model is obtained.
  • the model parameters of the reference signal identification model unchanged, and use the reference signal identification model to pair the virtual channel information (i.e. the fourth data) and the virtual reference signal (i.e. the fifth data) generated by the channel generation model to be trained. ) to identify the combined data, and determine the probability that the category of the current combined data is the category to which the above-mentioned real combined data belongs (that is, the second identification result). Furthermore, for the purpose of increasing the probability, the model parameters of the reference signal generation model to be trained can be adjusted so that the reference signal identification model cannot distinguish the difference between the combination data and the above-mentioned real combination data.
  • the above steps can be repeated. After multiple update iterations, when the reference signal identification model cannot identify whether the combined data is generated or real, that is, the reference signal identification model identifies the probability that the category of the combined data is the category to which the real combined data belongs. When it is greater than the third threshold, that is, the reference signal discrimination model and the reference signal generation model to be trained reach a stable state, and the training is completed. In this way, a trained reference signal generation model conditioned on the channel is obtained.
  • the data acquisition device can separately extract the trained reference signal generation model to obtain a channel-conditioned reference signal generation model for generating reference signal sample data associated with channel information sample data.
  • the generative model can be a single complete generative model, that is, the generative model can simultaneously generate A pair of reference signal sample data and channel information sample data with an associated relationship.
  • the generation model may not have independent input data, that is, the generation model may be directly used to generate reference signal sample data and channel information sample data.
  • the input data of the generating model may also be at least one of the following: noise, random number, channel type indication information, real reference signal, statistical information of real reference signal, real channel information, and statistical information of real channel information.
  • the output data of the generation model may be a set of data composed of reference signal sample data and channel information sample data.
  • the output data of the generative model may be the first reference signal sample data of size M*N*P shown in FIG. 9, and the first channel information of size I*J*K shown in FIG. 10 Three-dimensional data with a size of (M+I)*(N+J)*(P+K) composed of sample data.
  • the training process of the model is similar to the training process of the above-mentioned embodiment, and for the sake of brevity, details are not repeated here.
  • the embodiment of the present application provides a data acquisition method.
  • the data acquisition device can acquire the reference signal sample data and channel information sample data generated by the generation model; The signal and the real channel information are trained; in addition, the reference signal sample data and channel information sample data are used to train the channel estimation model.
  • the reference signal sample data and channel information sample data in this application are generated by a well-trained generative model, and the generative model is constructed based on a generative confrontation network, which requires only a small amount of real reference signal and real channel information. Realize the training of the generative model, avoiding the process of collecting a large number of real reference signals and real channel information and modeling the channel, and greatly reducing the difficulty of data acquisition and manual overhead.
  • Fig. 18 is a schematic diagram of the first structural composition of the data acquisition device provided by the embodiment of the present application. As shown in Fig. 18, the data acquisition device includes:
  • the acquiring unit 1801 is configured to acquire reference signal sample data and channel information sample data generated by a generative model; the generative model is obtained based on generative confrontation network, real reference signal, and real channel information training;
  • the reference signal sample data and channel information sample data are used to train a channel estimation model.
  • the reference signal sample data and the channel information sample data have an association relationship.
  • the real reference signal is received through a channel corresponding to the real channel information.
  • the number of the real reference signal and the real channel information is a preset number; the preset number is less than or equal to a first threshold.
  • the generative models include a first generative model and a second generative model
  • the obtaining unit 1801 is specifically configured to obtain first data generated by the first generation model; input the first data into the second generation model, and obtain second data generated by the second generation model ;
  • the first generation model is a reference signal generation model
  • the second generation model is a channel generation model
  • the first generation model is a channel generation model
  • the second generation model is a reference signal generation model
  • the reference signal generation model is used to generate the reference signal sample data
  • the channel generation model is used to generate the channel information sample data.
  • it also includes a first training unit configured to: acquire the third data output by the first generation model to be trained; input the third data into the first identification model, and output the third data through the first identification model An identification result; the first identification model is used to identify the probability that the category corresponding to the third data is the category to which the first real data belongs; the first real data has an association relationship with the first generation model; based on the According to the first identification result, adjust the model parameters of the first generation model to be trained to obtain the first generation model; the category corresponding to the data generated by the first generation model is the category to which the first real data belongs The probability of is greater than the second threshold.
  • the first generation model is a reference signal generation model, and the first generation model has no input data in the process of generating data, or the first generation model uses at least one of the following data as Input data:
  • Noise, random numbers, channel type indication information, true reference signal, and statistical information of the true reference signal are examples of the true reference signal.
  • the first generation model is a channel information generation model, and the first generation model has no input data in the process of generating data, or the first generation model uses at least one of the following data as Input data:
  • Noise, random numbers, channel type indication information, real channel information, and statistical information of real channel information are examples of real channel information.
  • the channel type indication information includes at least one of the following:
  • the first indication information is used to indicate the frequency band where the channel is located
  • the second indication information is used to indicate the radio frequency environment
  • the third indication information is used to indicate the wireless channel scenario.
  • it also includes a second training unit configured to input fourth data into the second generation model to be trained, and generate fifth data through the second generation model to be trained, the fourth data is the The data generated by the first generation model; the fifth data is input into the second identification model, and the second identification result is output through the second identification model; the second identification result is used to identify the category corresponding to the combined data as the first Probability of the category to which the second real data belongs; the combined data includes the fourth data and the fifth data; the second real data includes the real reference signal and the real channel information; based on the second identification As a result, the model parameters of the second generation model to be trained are adjusted to obtain the second generation model; the data formed by combining the data generated by the second generation model with the fourth data corresponds to The probability of the category and the category to which the second real data belongs is greater than a third threshold.
  • the second generation model is a channel information generation model
  • the fourth data is data generated by a reference signal generation model
  • the second generation model is a reference signal generation model
  • the fourth data is data generated by a channel information generation model
  • FIG. 19 is a schematic structural diagram of an electronic device 1900 provided by an embodiment of the present application.
  • the electronic device may be a third-party server, a terminal device, or a network device, which is not limited in this embodiment of the present application.
  • the electronic device 1900 shown in FIG. 19 includes a processor 1910, and the processor 1910 can call and run a computer program from a memory, so as to implement the method in the embodiment of the present application.
  • the electronic device 1900 may further include a memory 1920 .
  • the processor 1910 can invoke and run a computer program from the memory 1920, so as to implement the data acquisition method in the embodiment of the present application.
  • the memory 1920 may be an independent device independent of the processor 1910 , or may be integrated in the processor 1910 .
  • FIG. 20 is a schematic structural diagram of a chip according to an embodiment of the present application.
  • the chip 2000 shown in FIG. 20 includes a processor 2010, and the processor 2010 can call and run a computer program from a memory, so as to implement the method in the embodiment of the present application.
  • the chip 2000 may further include a memory 2020 .
  • the processor 2010 can invoke and run a computer program from the memory 2020, so as to implement the method in the embodiment of the present application.
  • the memory 2020 may be a separate device independent of the processor 2010 , or may be integrated in the processor 2010 .
  • the chip 2000 may also include an input interface 2030 .
  • the processor 2010 can control the input interface 2030 to communicate with other devices or chips, specifically, can obtain information or data sent by other devices or chips.
  • the chip 2000 may also include an output interface 2040 .
  • the processor 2010 can control the output interface 2040 to communicate with other devices or chips, specifically, can output information or data to other devices or chips.
  • the chip can be applied to the network device in the embodiment of the present application, and the chip can implement the corresponding processes implemented by the network device in the methods of the embodiment of the present application.
  • the chip can implement the corresponding processes implemented by the network device in the methods of the embodiment of the present application.
  • the chip can be applied to the mobile terminal/terminal device in the embodiments of the present application, and the chip can implement the corresponding processes implemented by the mobile terminal/terminal device in the various methods of the embodiments of the present application.
  • the chip can implement the corresponding processes implemented by the mobile terminal/terminal device in the various methods of the embodiments of the present application.
  • the chip can implement the corresponding processes implemented by the mobile terminal/terminal device in the various methods of the embodiments of the present application.
  • the chip can be applied to the mobile terminal/terminal device in the embodiments of the present application, and the chip can implement the corresponding processes implemented by the mobile terminal/terminal device in the various methods of the embodiments of the present application.
  • the chip mentioned in the embodiment of the present application may also be called a system-on-chip, a system-on-chip, a system-on-a-chip, or a system-on-a-chip.
  • the processor in the embodiment of the present application may be an integrated circuit chip, which has a signal processing capability.
  • each step of the above-mentioned method embodiments may be completed by an integrated logic circuit of hardware in a processor or instructions in the form of software.
  • the above-mentioned processor can be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), an off-the-shelf programmable gate array (Field Programmable Gate Array, FPGA) or other available Program logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field Programmable Gate Array
  • a general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.
  • the steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor.
  • the software module can be located in a mature storage medium in the field such as random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register.
  • the storage medium is located in the memory, and the processor reads the information in the memory, and completes the steps of the above method in combination with its hardware.
  • the memory in the embodiments of the present application may be a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memories.
  • the non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electronically programmable Erase Programmable Read-Only Memory (Electrically EPROM, EEPROM) or Flash.
  • the volatile memory can be Random Access Memory (RAM), which acts as external cache memory.
  • RAM Static Random Access Memory
  • SRAM Static Random Access Memory
  • DRAM Dynamic Random Access Memory
  • Synchronous Dynamic Random Access Memory Synchronous Dynamic Random Access Memory
  • SDRAM double data rate synchronous dynamic random access memory
  • Double Data Rate SDRAM, DDR SDRAM enhanced synchronous dynamic random access memory
  • Enhanced SDRAM, ESDRAM synchronous connection dynamic random access memory
  • Synchlink DRAM, SLDRAM Direct Memory Bus Random Access Memory
  • Direct Rambus RAM Direct Rambus RAM
  • the memory in the embodiment of the present application may also be a static random access memory (static RAM, SRAM), a dynamic random access memory (dynamic RAM, DRAM), Synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection Dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is, the memory in the embodiments of the present application is intended to include, but not be limited to, these and any other suitable types of memory.
  • the embodiment of the present application also provides a computer-readable storage medium for storing computer programs.
  • the computer-readable storage medium can be applied to the network device in the embodiment of the present application, and the computer program enables the computer to execute the corresponding process implemented by the electronic device in each method of the embodiment of the present application.
  • the computer program enables the computer to execute the corresponding process implemented by the electronic device in each method of the embodiment of the present application.
  • the embodiment of the present application also provides a computer program product, including computer program instructions.
  • the computer program product may be applied to the electronic device in the embodiment of the present application, and the computer program instructions cause the computer to execute the corresponding process implemented by the network device in each method of the embodiment of the present application.
  • the Let me repeat for the sake of brevity, the Let me repeat.
  • the embodiment of the present application also provides a computer program.
  • the computer program can be applied to the electronic device in the embodiment of the present application.
  • the computer program executes the corresponding process implemented by the network device in each method of the embodiment of the present application.
  • the network device For the sake of brevity , which will not be repeated here.
  • the disclosed systems, devices and methods may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the functions described above are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory,) ROM, random access memory (Random Access Memory, RAM), magnetic disk or optical disc, etc., which can store program codes. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Power Engineering (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请实施例提供一种数据获取方法、装置、设备、介质、芯片、产品及程序，该方法包括：获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。

Description

数据获取方法、装置、设备、介质、芯片、产品及程序 技术领域
本申请实施例涉及移动通信技术领域，具体涉及一种数据获取方法、装置、设备、介质、芯片、产品及程序。
背景技术
鉴于人工智能(Artificial Intelligence，AI)技术在计算机视觉、自然语言处理等方面取得了巨大的成功，通信领域开始尝试利用AI技术来寻求新的技术思路来解决传统方法受限的技术难题。
如何对通信领域中的AI模型进行训练，是本领域一直关注的问题。
发明内容
本申请实施例提供一种数据获取方法、装置、设备、介质、芯片、产品及程序。
本申请实施例提供一种数据获取方法，包括：
获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；
所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。
本申请实施例提供一种数据获取装置，包括：
获取单元，配置为获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；
所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。
本申请实施例提供一种电子设备，包括：存储器和处理器，
所述存储器存储有可在处理器上运行的计算机程序，
所述处理器执行所述程序时实现所述数据获取方法。
本申请实施例提供一种计算机存储介质，所述计算机存储介质存储有一个或者多个程序，所述一个或者多个程序可被一个或者多个处理器执行，以实现所述数据获取方法。
本申请实施例提供一种芯片，包括：处理器，用于从存储器中调用并运行计算机程序，使得安装有所述芯片的设备执行所述数据获取方法的步骤。
本申请实施例提供一种计算机程序产品，所述计算机程序产品包括计算机存储介质，所述计算机存储介质存储计算机程序，所述计算机程序包括能够由至少一个处理器执行的指令，当所述指令由所述至少一个处理器执行时实现所述数据获取方法的步骤。
本申请实施例提供一种计算机程序，所述计算机程序使得计算机执行所述数据获取方法。
本申请实施例提供一种数据获取方法，具体地，数据获取装置可以获取生成模型生成的参考信号样本数据和信道信息样本数据；其中，该生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；另外，参考信号样本数据和信道信息样本数据用于训练信道估计模型。可以看到，本申请中的参考信号样本数据和信道信息样本数据是通过训练好的生成模型生成的，而生成模型是基于生成对抗网络构建，仅需少量的真实参考信号和真实信道信息便可实现生成模型的训练，避免了采集大量的真实参考信号及真实信道信息以及对信道的建模过程，大大降低了数据获取的难度和人工开销。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1为本申请实施例提供的一种无线通信系统的通信流程示意图；
图2为本申请实施例提供的一种无线通信系统中信道估计及恢复过程示意图；
图3为相关技术提出的一种神经网络的结构示意图；
图4为相关技术提供的一种卷积神经网络的结构示意图；
图5为本申请实施例提供的一种利用AI实现信道估计与恢复的示意图；
图6为本申请实施例提供的一种数据获取方法的流程示意图一；
图7为本申请实施例提供的一种数据获取方法的流程示意图二；
图8为本申请实施例提供的一种参考信号生成模型和以参考信号生成模型为条件的信道生成模型的接口示意图；
图9为本申请实施例提供的一种参考信号数据结构示意图；
图10为本申请实施例提供的一种信道信息数据结构示意图；
图11为本申请实施例提供的一种信道生成模型和以信道为条件的参考信号生成模型的接口示意图；
图12为本申请实施例提供的一种参考信号生成模型的训练流程示意图；
图13为本申请实施例提供的一种生成对抗网络的网络结构示意图一；
图14为本申请实施例提供的一种信道生成模型的训练流程示意图；
图15为本申请实施例提供的一种生成对抗网络的网络结构示意图二；
图16为本申请实施例提供的一种以参考信号为条件的信道生成模型的训练流程示意图；
图17为本申请实施例提供的一种以信道为条件的参考信号生成模型的训练流程示意图。
图18是本申请实施例提供的一种数据获取装置的示意性结构图；
图19是本申请实施例提供的一种电子设备的示意性结构图；
图20是本申请实施例提供的一种芯片的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
为便于理解本申请实施例的技术方案，以下对本申请实施例的相关技术进行说明，以下相关技术作为可选方案与本申请实施例的技术方案可以进行任意结合，其均属于本申请实施例的保护范围。
图1为本申请实施例提供的一种无线通信系统的通信流程示意图，如图1所示，该无线通信系统可以包括发射端和接收端。
在发射端，发射机101对信源比特流进行信道编码、调制，获得调制数据；在调制数据中插入参考信号，插入的参考信号用于接收端的信道估计，最后形成发送信号，经过信道到达接收端。其中，发送信号经过信道发送到接收端的过程中会受到噪声的干扰。
在接收端，接收机102首先接收发送端传输的信号，得到接收信号，进而利用接收信号中的参考信号进行信道估计，得到信道状态信息(Channel State Information，CSI)。接收端通过反馈链路将CSI反馈给发送端，供发射机调整信道编码、调制、预编码等方式，最后，接收机通过对接收信号进行解调以及信道解码等步骤，获得最终的恢复比特流。
需要说明的是，图1是对无线通信系统的通信流程进行了简单的示意，无线通信系统中还有其他未列举的如资源映射、预编码、干扰消除、CSI测量等模块，这些模块也都是单独设计实现，然后各个独立模块整合后可构成一个完整的无线通信系统。
还需要说明的是，上述无线通信系统可以是长期演进(Long Term Evolution，LTE)系统、LTE时分双工(Time Division Duplex，TDD)、通用移动通信系统(Universal Mobile Telecommunication System，UMTS)、物联网(Internet of Things，IoT)系统、窄带物联网(Narrow Band Internet of Things，NB-IoT)系统、增强的机器类型通信(enhanced Machine-Type Communications，eMTC)系统、5G通信系统(也称为新无线(New Radio，NR)通信系统)，或未来的通信系统(例如6G通信系统)等。
由于无线信道环境的复杂性和时变性，在无线通信系统中，接收机针对无线信道的估计及恢复会直接影响数据的接收情况，从而影响通信系统的性能。
图2为本申请实施例提供的一种无线通信系统中信道估计及恢复过程示意图，如图2所示，如(a)所示发射机在时频资源上发送的信号，除了信息数据符号外，还会发送一系列接收机已知的特定导频符号(即参考信号符号)。其中，参考信号可以是信道状态信息参考信号(Channel State  Information-Reference Signal，CSI-RS)信号、解调参考信号(DeModulation Reference Signal，DMRS)信号等。
发射机发送的信号经过信道传输到接收机，如(b)所示，接收机接收的数据符号和参考信号符号携带有噪音(即携带噪音的数据符号和参考信号符号)，接收机可以基于携带噪音的数据符号和参考信号符号进行信道估计。针对信道估计阶段，如(c)所示，接收机可根据已知参考信号符号的取值与接收导频符号的取值，利用最小二乘法(Least Squares，LS)或者最小均方误差(Minimum Mean Square Error，MMSE)等方法估计出该参考信号所处时频位置上的信道信息。然后接收机可以基于该信道信息进行信道恢复，针对信道恢复阶段，如(d)所示，接收机根据参考信号符号位置上估计出的信道信息，并利用插值算法恢复出全时频资源上的信道信息，用于后续的信道信息反馈或数据恢复等。
从(c)中可以看出，信道已估计/恢复的资源为参考信号符号的位置上的资源。从(d)中可以看出，信道已估计/恢复的资源为参考信号符号和数据符号的位置上的资源。
近年来，以神经网络为代表的人工智能研究在很多领域都取得了非常大的成果，其也将在未来很长一段时间内在人们的生产生活中起到重要的作用。
图3为相关技术提出的一种神经网络的结构示意图，如图3所示，神经网络的结构可以包括：输入层，隐藏层和输出层，如图3所示，输入层负责接收数据，隐藏层对数据的处理，最后的结果在输出层产生。在这其中，各个节点代表一个处理单元，可以认为是模拟了一个神经元，多个神经元组成一层神经网络，多层的信息传递与处理构造出一个整体的神经网络。
随着神经网络研究的不断发展，近年来又提出了神经网络深度学习算法，较多的隐层被引入，通过多隐层的神经网络逐层训练进行特征学习，极大地提升了神经网络的学习和处理能力，并在模式识别、信号处理、优化组合、异常探测等方面广泛被应用。
同样，随着深度学习的发展，卷积神经网络(Convolutional Neural Networks，CNN)也被进一步研究。
图4为相关技术提供的一种卷积神经网络的结构示意图，如图4所示，卷积神经网络的结构可以包括：输入层、多个卷积层、多个池化层、全连接层及输出层。通过卷积层和池化层的引入，有效地控制了网络参数的剧增，限制了参数的个数并挖掘了局部结构的特点，提高了算法的鲁棒性。
实际应用中，可以利用AI实现信道的估计与恢复。具体地，图5为本申请实施例提供的一种利用AI实现信道估计与恢复的示意图，如图5所示，可以通过神经网络构建基于AI的信道估计模型(或称为信道恢复模型)，利用信道估计模型对信道进行估计与恢复。其中，基于AI的信道估计模型51的输入数据为参考信号，输出数据为传输该参考信号的信道所对应的信道信息(即信道估计结果)。这里需要注意的是，基于AI的信道估计模型51的输入信息除了参考信号外还可以增加其他辅助信息用于提升基于AI的信道估计模型51性能，例如，这些其他辅助信息可以是对参考信号的特征提取得到的信息、能量水平、时延特征、噪声特征等。
需要说明的是，输入基于AI的信道估计模型/信道恢复模型51的参考信号，至少包括接收信号中的参考信号，即携带噪音的参考信号。
显然，基于AI的信道估计模型可以基于参考信号，以及对应的信道信息训练得到。也就是说，基于AI的信道估计模型的训练数据集包括参考信号，以及与该参考信号对应的信道信息。通常情况下，获得上述训练数据集最直接的方式是对实际的无线信道进行采集，例如通过配对的信号发射机和信号接收机获取接收的参考信号以及该参考信号对应的信道信息，或者通过特定的接收机采集第三方发射机(例如蜂窝网络基站)的信号从而获取接收机所接收的参考信号以及该参考信号对应的信道信息。另外一种方法是利用信道的数学模型所建立的仿真平台，生成大量的训练样本数据。
目前，基于AI的信道估计模型对训练数据集(包括参考信号和其对应的信道信息)具有极大的依赖和需求。可以说，训练数据集是决定该类方案性能增益的关键。
然而，随着无线通信系统的发展，无线通信频段逐渐由低频迈向高频，并且逐步走向更加复杂的空天地海等特殊环境，在应用范围上向人机交互、物联网交互、工业应用、特种应用等更多场景扩展，使得当前无线通信系统所需要面对的无线信道环境越来越复杂。因此，在上述复杂的无线信道环境下，对于接收参考信号以及信道信息的采集十分困难，这里的困难既有技术层面的困难，也有操作层面的困难。与此同时，对于上述复杂信道的数学建模也面临巨大挑战，频段、环境、场景的复杂性会知道导致信道建模的复杂性，非线性的信道特征以及难以拟合的信道传播特性会对传统数学建模来研究信道的方式带来困难与挑战。例如，在目前高频信道建模还是一个亟待解决的问题，更进一步地，复杂场景和应用环境下的实际信道环境建模与理想信道建模之间的差异，在未来无线 通信研究中还将继续随着信道环境的复杂程度极具增加。
事实上，构建基于AI的信道估计模型对于训练数据集的依赖程度极高，往往需要几千、几万、几十万甚至更大规模的数据。在复杂的无线信道环境下，获取信道估计模型所需的大量的样本数据显然十分困难。一方面传统的实采或建模的方法会在可实现性、可靠性方面存在较大的问题。另一方面，实采如此规模的训练样本数据的人工成本代价极高。
综上所述，如何获取、构建有效的信道估计模型的训练数据集，进而对信道估计模型进行训练，是一个亟待解决的关键问题。
基于此，本申请实施例提供一种数据获取方法，具体地，数据获取装置可以获取生成模型生成的参考信号样本数据和信道信息样本数据；其中，该生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；另外，参考信号样本数据和信道信息样本数据用于训练信道估计模型。可以看到，本申请中的参考信号样本数据和信道信息样本数据是通过训练好的生成模型生成的，而生成模型是基于生成对抗网络构建，仅需少量的真实参考信号和真实信道信息便可实现生成模型的训练，避免了采集大量的真实参考信号及真实信道信息以及对信道的建模过程，大大降低了数据获取的难度和人工开销。
需要说明的是，本申请实施例提供的数据获取方法，可以应用于本申请实施例提供的数据获取装置。该数据获取装置可以通过软件或者硬件的方式集成在电子设备中，该可以是服务器、个人计算机、工业计算机等，另外，电子设备还可以是具备通信功能的网络设备或终端设备等，本申请实施例对此不做限定。
其中，网络设备可以是LTE系统中的演进型基站(Evolutional Node B，eNB或eNodeB)，或者是下一代无线接入网(Next Generation Radio Access Network，NG RAN)设备，或者是NR系统中的基站(gNB)，或者是云无线接入网络(Cloud Radio Access Network，CRAN)中的无线控制器，或者该网络设备可以为中继站、接入点、车载设备、可穿戴设备、集线器、交换机、网桥、路由器，或者未来演进的公共陆地移动网络(Public Land Mobile Network，PLMN)中的网络设备等。
终端设备110可以是任意终端设备，其包括但不限于与网络设备120或其它终端设备采用有线或者无线连接的终端设备。例如，所述终端设备110可以指接入终端、用户设备(User Equipment，UE)、用户单元、用户站、移动站、移动台、远方站、远程终端、移动设备、用户终端、终端、无线通信设备、用户代理或用户装置。接入终端可以是蜂窝电话、无绳电话、会话启动协议(Session Initiation Protocol，SIP)电话、IoT设备、卫星手持终端、无线本地环路(Wireless Local Loop，WLL)站、个人数字处理(Personal Digital Assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、车载设备、可穿戴设备、5G网络中的终端设备或者未来演进网络中的终端设备等。
为便于理解本申请实施例的技术方案，以下通过具体实施例详述本申请的技术方案。以上提供的技术作为可选方案与本申请实施例的技术方案可以进行任意结合，其均属于本申请实施例的保护范围。本申请实施例包括以下内容中的至少部分内容。
图6是本申请实施例提供的数据获取方法的流程示意图一，如图6所示，该方法包括以下内容。
步骤610、获取生成模型生成的参考信号样本数据和信道信息样本数据；生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；参考信号样本数据和信道信息样本数据用于训练信道估计模型。
应理解，这里的生成模型是指预先训练好的模型。其中，生成模型可以基于生成对抗网络、真实参考信号、以及真实信道信息训练得到的。
生成对抗网络(Generative Adversarial Networks，GAN)是一种深度学习模型，在复杂的数据分布上无监督学习最具前景的方法之一。生成对抗网络通过(至少)两个模型：生成模型(Generative Model)和鉴别模型(Discriminative Model)之间的互相博弈学习产生符合预期的输出。其中，生成模型用于生成虚拟数据，而鉴别模型用于鉴别生成模型所生成的虚拟数据的真假(或者说用于鉴别该虚拟数据是生成模型生成的数据，还是来自真实数据)。实际应用中，可以对鉴别模型和生成模型进行交替训练，使得训练好的生成模型所生成的数据可以骗过鉴别模型。也就是说，训练好的生成模型所生成的虚拟数据，鉴别模型是无法鉴别出该虚拟数据是生成模型生成的数据还是真实的数据。可见，训练好的生成模型生成的虚拟数据可以媲美真实数据。
在本申请实施例中的生成模型可以基于真实参考信号以及真实信道信息训练得到。本申请实施例中的生成模型具体用于生成虚拟的参考信号样本数据和虚拟的信道信息样本数据。
可以理解的是，本申请实施例中的生成模型，其所生成的参考信号样本数据和信道信息样本数 据可以十分接近真实参考信号和信道信息。因此，通过生成模型可以生成大量的参考信号样本数据和信道信息样本数据，以构成训练信道估计模型的训练数据集。
本申请实施例中，用于训练上述生成模型的真实参考信号，可以至少包括接收机接收到的参考信号(也可以称为接收参考信号)，即携带噪音的参考信号。这里，真实参考信号和真实信道信息之间具有关联关系，一个真实参考信号可以对应一个真实信道信息。为了方便描述，下文中的参考信号均指接收参考信号。
可选地，真实参考信号是通过真实信道信息对应的信道接收得到。也就是说，接收机在该真实信道信息对应的信道上接收到上述真实参考信号。
可选地，用于生成模型训练的真实参考信号和真实信道信息的数量均为预设数量；预设数量小于或等于第一阈值。其中，第一阈值可以是50或100，本申请实施例对此不做限定。
可以看出，用于训练本申请提供的生成模型的真实参考信号和真实信道信息的数量，远远小于相关技术中训练信道估计模型所需的成千上万的真实参考信号和真实信道信息。也就是说，本申请实施例可以仅依赖少量的真实参考信号和真实信道信息，来构建生成模型。这样，本申请实施例提供的生成模型可以生成大量且接近真实的参考信号样本数据和信道信息样本数据，以用于信道估计模型的训练，避免了采集大量的真实参考信号及真实信道信息以及对信道的建模过程，大大降低了数据获取的难度和人工开销。
需要说明的是，上述多个真实参考信号和真实信道信息可以是在多种频段、射频环境、以及无线信道场景中获取到的真实数据。这样，基于不同频段、射频环境、以及无线信道场景的真实参考信号和真实信道信息训练得到的生成模型，也可以生成种频段、射频环境、以及无线信道场景对应的参考信号样本数据，以及信道信息样本数据。如此，保证了参考信号样本数据和信道信息样本数据的多样性。
可选地，由于真实参考信号和真实信道信息具有关联关系，对应的，生成模型生成的参考信号样本数据和信道信息样本数据之间也具有关联关系。参考信号样本数据和信道信息样本数据之间可以一一对应，即一个参考信号样本数据对应一个信道信息样本数据。这样，具有关联关系的参考信号样本数据和信道信息样本数据可以构成虚拟的参考信号-信道信息数据对，用于信道估计模型的训练。
由此可见，本申请实施例提供的数据获取方法中，数据获取装置可以获取生成模型生成的参考信号样本数据和信道信息样本数据；其中，该生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；另外，参考信号样本数据和信道信息样本数据用于训练信道估计模型。可以看到，本申请中的参考信号样本数据和信道信息样本数据是通过训练好的生成模型生成的，而生成模型是基于生成对抗网络构建，仅需少量的真实参考信号和真实信道信息便可实现生成模型的训练，避免了采集大量的真实参考信号及真实信道信息以及对信道的建模过程，大大降低了数据获取的难度和人工开销。
以下详细介绍如何利用生成模型生成参考信号样本数据和信道信息样本数据。
在一些实施例中，生成模型可以包括参考信号生成模型，以及信道生成模型。具体地，参考信号生成模型用于生成参考信号样本数据，信道生成模型用于生成信道信息样本数据。也就是说，本申请实施例提供的生成模型，可以利用两个独立的生成模型，分别生成具有关联关系参考信号样本数据和信道信息样本数据。
具体地，参考图7所示的流程示意图，步骤610中获取生成模型生成的参考信号样本数据和信道信息样本数据，可以通过以下步骤实现：
步骤6101、获取第一生成模型生成的第一数据；
步骤6102、将第一数据输入至第二生成模型中，获取第二生成模型生成的第二数据。
在一种可能的实现方式中，第一生成模型可以为参考信号生成模型，对应的，第二生成模型可以为信道生成模型。
在该实现方式中，信道生成模型也可以称为以参考信号为条件的信道生成模型。
图8示出了参考信号生成模型和以参考信号生成模型为条件的信道生成模型的接口示意图。其中，参考信号生成模型的输出端，可以与以参考信号为条件的信道生成模型的输入端连接。
也就是说，数据获取装置可以先获取参考信号生成模型生成的第一参考信号样本数据(即第一数据)，进而，数据获取装置可以将该第一参考信号样本数据输入至以参考信号为条件的信道生成模型中。这样，以参考信号为条件的信道生成模型可以基于输入的第一参考信号样本数据生成与该第一参考信号样本数据具有关联关系的第一信道信息样本数据(即第二数据)。
可选地，参考信号生成模型可以无输入数据，也就是说，数据获取装置可以直接利用参考信号生成模型生成第一参考信号样本数据。
可选地，参考信号生成模型还可以以下中的至少一项数据作为输入数据：
噪声、随机数、信道类型指示信息、真实参考信号、以及真实参考信号的统计信息。
其中，噪声可以是来自于真实环境的噪声数据，也可以是人工生成的噪声数据，本申请实施例对此不做限定。
另外，随机数可以是随机数序列，也可以是伪随机数序列，本申请实施例对此不做限定。
需要说明的是，作为参考信号生成模型输入数据的噪声、随机数的数据格式可以是一维向量、二维矩阵、或者更高维度的数据，本申请实施例对此不做限定。另外，噪声和随机数的数据格式可以是提前约定好的，或者是和生成的参考信号样本数据的数据格式一致，本申请实施例对此不做限定。
本申请实施例中的信道类型指示信息可以包括以下中的至少一项：
第一指示信息，用于指示信道所处的频段；
第二指示信息，用于指示射频环境；
第三指示信息，用于指示无线信道场景。
示例性的，第一指示信息可以指示当前处于的是高频还是低频。第二指示信息可以指示当前的射频环境是室内、室外、密集小区、或者空旷外场环境。第三指示信息可以指示当前无线信道场景是物联网场景，还是工业场景等。
另外，真实参考信号可以是实际采集的，或者是通过数学建模方式获取的参考信号。真实参考信号的统计信息包括但不限于真实参考信号的最大功率谱密度，平均功率谱密度等。
可以理解的是，参考信号生成模型可以基于上述输入数据生成一种或者多种频段、射频环境、无线信道场景下的第一参考信号样本数据，以保证生成的第一参考信号样本数据的多样性和丰富性。
本申请实施例中，参考图9所示的参考信号数据结构示意图，单个的第一参考信号样本数据可以由大小为M*N*P的三维数据构成，其中，M、N和P的取值可以相等也可以不相等，本申请实施例对此不做限定。此外，也可以将M*N*P的三维数据合成为1*(M*N*P)大小或者(M*N*P)*1大小的一维向量，本申请实施例对参考信号样本数据的数据格式不做限定。
图9所示的第一参考信号样本数据的三个维度可以分别代表频域维度、时域维度和天线对域维度。其中，频域维度的每个单位对应1个子载波，时域维度的每个单位对应1个时域符号，天线对维度的每个单位对应1个天线对。
需要说明的是，通常情况下，参考信号可以通过复数来呈现。因此，上述参考信号生成模型输出的第一参考信号样本数据可以在上述三维数据的基础上额外增加一个维度(即第四维度)，该第四维度可以表征第一参考信号样本数据的实部和虚部。但是，本申请实施例中的第一参考信号样本数据并不局限在四维以下，第一参考信号样本数据还可以包括更多维度的信息。
参考图8所示，本申请实施例中以参考信号为条件的信道生成模型，其输入可以是第一参考信号样本数据，其输出可以是与第一参考信号对应的信道信息，即第一信道信息样本数据。
可选地，与第一参考信号样本数据的数据格式类似，第一信道信息样本数据可以由频域、时域、天线对域三个维度的数据构成。示例性的，参考图10所示，第一信道信息样本数据可以由大小为I*J*K的三维数据构成。其中，I、J和K的取值可以相等也可以不相等，本申请实施例对此不做限定。此外，也可以将I*J*K的三维数据合成为1*(I*J*K)大小或者(I*J*K)*1大小的一维向量，本申请实施例对信道信息样本数据的数据格式不做限定。
可选地，I的取值可以与N的取值相同，J的取值可以与M的取值相同，P的取值可以与K的取值相同。
需要说明的是，信道信息也可以通过复数来呈现。因此，上述以参考信号为条件的信道生成模型输出的第一信道信息样本数据可以在上述三维数据的基础上额外增加一个维度(即第四维度)，该第四维度可以表征第一信道信息样本数据的实部和虚部。但是，本申请实施例中的第一信道信息样本数据并不局限在四维以下，第一信道信息样本数据还可以包括更多维度的信息。
还需要说明的是，以参考信号为条件的信道生成模型输出的第一信道信息样本数据，可以是原始信道信息通过数学变换后得到的信道特征信息，例如通过SVD分解得到的信道特征向量信息，可以是单流的信道特征向量信息，也可以是多流的信道特征向量信息，例如2流、4流、8流信道特征向量信息，本申请实施例对此不做限定。
在本申请实施例中，以参考信号为条件的信道生成模型，输入数据可以是一种或者多种频段、 射频环境、无线信道场景下的第一参考信号样本数据。因此，上述以参考信号为条件的信道生成模型对应的也可以生成一种或者多种频段、射频环境、无线信道场景下的第一信道信息样本数据。
如此，在本实现方式中，可以先通过参考信号生成模型生成大量虚拟的第一参考信号样本数据，进而利用所生成的第一参考信号样本数据，通过以参考信号为条件的信道生成模型生成各第一参考信号样本数据对应的第一信道信息样本数据，构成具有关联关系的多对参考信号-信道信息样本数据，以用于信道估计模型的训练。
在另一种实现方式中，第一生成模型可以为信道生成模型，对应的，第二生成模型可以为参考信号生成模型。
在该实现方式中，参考信号生成模型也可以称为以信道为条件的参考信号生成模型。
图11示出了信道生成模型和以信道为条件的参考信号生成模型的接口示意图。其中，信道生成模型的输出端，可以与以信道为条件的参考信号生成模型的输入端连接。
也就是说，数据获取装置可以先获取信道生成模型生成的第二信道信息样本数据(即第一数据)，进而，数据获取装置可以将该第二信道信息样本数据输入至以信道为条件的参考信号生成模型中。这样，以信道为条件的参考信号生成模型可以基于输入的第二信道信息样本数据生成与该第二信道信息样本数据具有关联关系的第二参考信号样本数据(即第二数据)。
可选地，信道生成模型可以无输入数据，也就是说，数据获取装置可以直接利用信道生成模型生成第二信道信息样本数据。
可选地，信道生成模型可以以下中的至少一项数据作为输入数据：
噪声、随机数、信道类型指示信息、真实信道信息、以及真实信道信息的统计信息。
其中，噪声、随机数、信道类型指示信息均与上述实施例中的描述类似，为了简洁，此处不再赘述。
此外，真实信道信息可以是实际采集的，或者是通过数学建模方式获取的信道信息。真实信道信息的统计信息包括但不限于真实信道信息的功率谱密度，平均功率谱密度等。
可以理解的是，信道生成模型可以基于上述输入数据生成一种或者多种频段、射频环境、无线信道场景下的第二信道信息样本数据，以保证生成的第二信道信息样本数据的多样性和丰富性。
本申请实施例中，参考图10所示，单个的第二信道信息样本数据由大小为I*J*K的时域、频域、天线对域的三维数据构成。其中，I、J和K的取值可以相等也可以不相等，本申请实施例对此不做限定。此外，也可以将I*J*K的三维数据合成为1*(I*J*K)大小或者(I*J*K)*1大小的一维向量，本申请实施例对信道信息样本数据的数据格式不做限定。
需要说明的是，通常情况下，信道信息可以通过复数来呈现。因此，上述信道生成模型输出的第二信道信息样本数据可以在上述三维数据的基础上额外增加一个维度(即第四维度)，该第四维度可以表征第二信道信息样本数据的实部和虚部。但是，本申请实施例中的第二信道信息样本数据并不局限在四维以下，第二信道信息样本数据还可以包括更多维度的信息。
参考图11所示，本申请实施例中以信道为条件的参考信号生成模型，其输入可以是第二信道信息样本数据，其输出可以是与第二信道信息样本数据对应的参考信号，即第二参考信号样本数据。
可选地，参考图9所示，与第二信道信息样本数据的数据格式类似，第二参考信号样本数据可以由大小为M*N*P的频域、时域、天线对域三个维度的数据构成。其中，M、N和P的取值可以相等也可以不相等，本申请实施例对此不做限定。此外，也可以将M*N*P的三维数据合成为1*(M*N*P)大小或者(M*N*P)*1大小的一维向量，本申请实施例对参考信号样本数据的数据格式不做限定。
可选地，N的取值可以与I的取值相同，M的取值可以与J的取值相同，K的取值可以与P的取值相同。
需要说明的是，上述以信道为条件的参考信号生成模型输出的第二参考信号样本数据可以在上述三维数据的基础上额外增加一个维度(即第四维度)，该第四维度可以表征第二参考信号样本数据的实部和虚部。但是，本申请实施例中的第二参考信号样本数据并不局限在四维以下，第二参考信号样本数据还可以包括更多维度的信息。
在本申请实施例中，以信道为条件的参考信号生成模型，输入数据可以是一种或者多种频段、射频环境、无线信道场景下的第二信道信息样本数据。因此，上述以信道为条件的参考信号生成模型对应的也可以生成一种或者多种频段、射频环境、无线信道场景下的第二参考信号样本数据。
如此，在本实现方式中，可以先通过信道生成模型生成大量虚拟的第二信道信息样本数据，进而利用所生成的第二信道信息样本数据，通过以信道为条件的参考信号生成模型生成各第二信道信息样本数据对应的第二参考信号样本数据，构成具有关联关系的多对参考信号-信道信息样本数据， 以用于信道评估模型的训练。
下面详细介绍第一生成模型的训练过程。
在本申请实施例中，第一生成模型的训练过程可以包括以下步骤：
步骤a1、获取待训练第一生成模型输出的第三数据；
步骤a2、将第三数据输入第一鉴别模型，通过第一鉴别模型输出第一鉴别结果；第一鉴别模型用于鉴别第三数据对应的类别为第一真实数据所属类别的概率；第一真实数据与第一生成模型具有关联关系；
步骤a3、基于第一鉴别结果，对待训练第一生成模型的模型参数进行调整，得到第一生成模型；第一生成模型生成的数据对应的类别为第一真实数据所属类别的概率大于第二阈值。
应理解，对于第一生成模型的训练，需要同时构建第一生成模型和第一鉴别模型两部分。其中，第一生成模型用于生成所需要的数据，第一鉴别模型用于鉴别第一生成模型生成的数据的真假(或者说用于鉴别该虚拟数据是生成模型生成的数据，还是来自真实数据)。通过对第一生成模型和第一鉴别模型的交替训练，使得训练好的第一生成模型所生成的数据可以骗过第一鉴别模型，即第一鉴别模型无法区分第一生成模型生成的数据到底来自真实的数据还是第一生成模型所生成的。
其中，第一真实数据与第一生成模型具有关联关系，若第一生成模型为参考信号生成模型，则第一真实数据为真实参考信号，若第一生成模型为信道生成模型，则第一真实数据为真实信道信息。
在一种可能的实现方式中，第一生成模型可以为图8中的参考信号生成模型，对应的第一鉴别模型为参考信号鉴别模型，且第一真实数据为真实参考信号。具体地，该参考信号生成模型的训练过程可以参考图12所示。
首先构建待训练参考信号生成模型和待训练参考信号鉴别模型。其中，待训练参考信号生成模型和待训练参考信号鉴别模型可采用全连接网络、卷积神经网络、残差网络、自注意力机制网络中的一种或多种网络结构组成。
示例性的，参考图13所示，待训练参考信号生成模型和待训练参考信号鉴别模型可以由多个全连接层组成。不同的全连接层对应不同的维度，以提取不同的数据特征。
本申请实施例中，与上述实施例中参考信号生成模型的输入相同，待训练参考信号生成模型可以没有独立的输入，也可以将噪声、随机数、信道类型指示信息、真实参考信号、以及真实参考信号的统计信息中的至少一项作为输入数据。
另外，待训练参考信号鉴别模型的输入可以是待训练参数信号生成模型生成的虚拟参考信号(即第三数据)，以及真实参考信号。
具体地，在训练时，可以先保持待训练参考信号生成模型的模型参数不变，通过待训练参考信号鉴别模型对待训练参考信号生成模型生成的虚拟参考信号(即第三数据)进行鉴别，确定当前虚拟参考信号的类别是真实参考信号类别的概率(即第一鉴别结果)。
进而，以降低该概率为目的，对待训练参考信号鉴别模型的模型参数进行调整，使得待训练参考信号鉴别模型尽可能地区分参考信号的真假，得到训练好的参考信号鉴别模型。
进一步，如图12所示，保持参考信号鉴别模型的模型参数不变，通过参考信号鉴别模型对待训练参考信号生成模型生成的虚拟参考信号(即第三数据)进行鉴别，确定当前生成的虚拟参考信号的类别是真实参考信号所属类别的概率(即第一鉴别结果)。进而，可以以提高该概率为目的，调整待训练参考信号生成模型的模型参数，使得参考信号鉴别模型无法区分待训练参考信号生成模型生成的虚拟参考信号与真实参考信号的差别。
可以重复上述步骤，经过多次更新迭代后，当参考信号鉴别模型无法鉴别出待训练参考信号生成模型生成的虚拟参考信号到底是生成的还是真实的，即参考信号鉴别模型鉴别出待训练参考信号生成模型生成的虚拟参考信号对应的类别为真实参考信号所属类别的概率大于第二阈值时，也即参考信号鉴别模型和待训练参考信号生成模型达到稳定状态，完成训练，得到训练好的参考信号生成模型。
这样，数据获取装置可以单独提取出训练好的参考信号生成模型，用于参考信号样本数据的生成。
在另一种可能的实现方式中，第一生成模型可以为图11中的信道生成模型，对应的第一鉴别模型为信道鉴别模型，并且第一真实数据为真实信道信息。具体地，该信道生成模型的训练过程可以参考图14所示。
首先构建待训练信道生成模型和待训练信道鉴别模型。其中，待训练信道生成模型和待训练信道鉴别模型可采用全连接网络、卷积神经网络、残差网络、自注意力机制网络中的一种或多种网络 结构组成。
示例性的，参考图15所示，待训练信道生成模型和待训练信道鉴别模型可以由多个全连接层组成。不同的全连接层对应不同的维度，以提取不同的数据特征。
本申请实施例中，与上述实施例中信道生成模型的输入相同，待训练信道生成模型可以没有独立的输入，也可以将噪声、随机数、信道类型指示信息、真实信道信息、以及真实信道信息的统计信息中的至少一项作为输入数据。
另外，待训练信道鉴别模型的输入可以是待训练信道生成模型生成的虚拟信道信息(即第三数据)，以及真实信道信息。
具体地，在训练时，可以先保持待训练信道生成模型的模型参数不变，通过待训练信道鉴别模型对待训练信道生成模型生成的虚拟信道信息(即第三数据)进行鉴别，确定当前虚拟信道信息的类别是真实信道信息所属类别的概率(即第一鉴别结果)。进而，以降低该概率为目的，对待训练信道鉴别模型的模型参数进行调整，使得待训练信道鉴别模型尽可能地区分信道信息的真假，得到训练好的参考信号鉴别模型。
进一步，如图14所示，保持信道鉴别模型的模型参数不变，通过信道鉴别模型对待训练信道生成模型生成的虚拟信道信息(即第三数据)进行鉴别，确定当前生成的虚拟信道信息的类别是真实信道信息所属类别的概率(即第一鉴别结果)。进而，可以以提高该概率为目的，调整待训练信道生成模型的模型参数，使得信道鉴别模型无法区分待训练信道生成模型生成的虚拟信道信息与真实信道信息的差别。
可以重复上述步骤，经过多次更新迭代后，当信道鉴别模型无法鉴别出待训练信道生成模型生成的虚拟信道信息到底是生成的还是真实的，即信道鉴别模型鉴别出待训练信道生成模型生成的虚拟信道信息对应的类别为真实信道信息所属类别的概率大于第二阈值时，也即信道鉴别模型和待训练信道生成模型达到稳定状态，完成训练，得到训练好的信道生成模型。
这样，数据获取装置可以单独提取出训练好的信道生成模型，用于信道信息样本数据的生成。
下面详细介绍第二生成模型的训练过程。
在本申请实施例中，第二生成模型的训练过程可以包括以下步骤：
步骤b1、将第四数据输入至待训练第二生成模型中，通过待训练第二生成模型生成第五数据，第四数据为第一生成模型生成的数据；
步骤b2、将第五数据输入第二鉴别模型，通过第二鉴别模型输出第二鉴别结果；第二鉴别结果用于鉴别组合数据对应的类别为第二真实数据所属类别的概率；组合数据包括第四数据和第五数据；第二真实数据包括真实参考信号和真实信道信息；
步骤b3、基于第二鉴别结果，对待训练第二生成模型的模型参数进行调整，得到第二生成模型；第二生成模型生成的数据与第四数据组合后所形成的数据，其对应的类别与第二真实数据所属类别的概率大于第三阈值。
应理解，与第一生成模型的训练过程类似，对于第二生成模型的训练，也需要同时构建第二生成模型和第二鉴别模型两部分。具体地，第二生成模型用于生成所需要的数据，第二鉴别模型用于鉴别第二生成模型生成的数据的真假(或者说用于鉴别该虚拟数据是生成模型生成的数据，还是来自真实数据)。通过对第二生成模型和第二鉴别模型的交替训练，使得训练好的第二生成模型所生成的数据可以骗过第一鉴别模型，即第二鉴别模型无法区分第二生成模型生成的数据到底来自真实的数据还是第二生成模型所生成的。
在一种可能的实现方式中，第二生成模型可以是图8所示的以参考信号为条件的信道生成模型。图16示出了以参考信号为条件的信道生成模型的训练过程。
具体地，可以先构建待训练信道生成模型和待训练信道鉴别模型。其中，待训练信道生成模型和待训练信道鉴别模型可采用全连接网络、卷积神经网络、残差网络、自注意力机制网络中的一种或多种网络结构组成。示例性的，参考图15所示，待训练信道生成模型和待训练信道鉴别模型可以由多个全连接层组成。不同的全连接层对应不同的维度，以提取不同的数据特征。
在本实现方式中，如图16所示，待训练信道生成模型的输入数据(即第四数据)可以为虚拟参考信号。这里，虚拟参考信号可以是已经训练好的参考信号生成模型生成的参考信号样本数据。可以理解的是，将虚拟参考信号作为本实现方式中待训练信道生成模型的输入数据，这样，待训练信道生成模型可以生成与虚拟参考信号具有关联关系的信道信息，从而得到对应的参考信号-信道信息样本数据对。
如图16所示，本实现方式中的待训练信道鉴别模型的输入可以包括虚拟参考信号(即第四数据)、 待训练信道生成模型生成的虚拟信道信息(即第五数据)、真实参考信号、以及真实信道信息。
可选地，在对待训练信道鉴别模型进行输入时，可以将虚拟参考信号(即第四数据)和虚拟信道信息(即第五数据)进行联合输入，即将虚拟参考信号和虚拟信道信息组合为一组合数据输入。示例性的，可以将图9所示的大小为M*N*P的第一参考信号样本数据，和图10所示的大小为I*J*K的第一信道信息样本数据组合为(M+I)*(N+J)*(P+K)的三维数据，并将该(M+I)*(N+J)*(P+K)的三维数据输入待训练信道鉴别模型。
同样的，在对待训练信道鉴别模型进行输入时，可以将真实参考信号和真实信道信息进行联合输入，即将真实参考信号和真实信道信息组成为真实组合数据进行输入。
基于此，在训练时，可以先保持待训练信道生成模型的模型参数不变，通过待训练信道鉴别模型对虚拟参考信号(即第四数据)和待训练信道生成模型生成的虚拟信道信息(即第五数据)组成的组合数据进行鉴别，确定该组合数据对应的类别是真实组合数据所属类别的概率(即第二鉴别结果)。进而，以降低该概率为目的，对待训练信道鉴别模型的模型参数进行调整，使得待训练信道鉴别模型尽可能地区分组合数据的真假，得到训练好的参考信号鉴别模型。
进一步，如图16所示，保持信道鉴别模型的模型参数不变，通过信道鉴别模型对虚拟参考信号(即第四数据)和待训练信道生成模型生成的虚拟信道信息(即第五数据)组成的组合数据进行鉴别，确定当前组合数据的类别是上述真实组合数据所属类别的概率(即第二鉴别结果)。进而，可以以提高该概率为目的，调整待训练信道生成模型的模型参数，使得信道鉴别模型无法区分组合数据与上述真实组合数据的差别。
可以重复上述步骤，经过多次更新迭代后，当信道鉴别模型无法鉴别出组合数据到底是生成的还是真实的，即信道鉴别模型鉴别出组合数据应的类别为真实组合数据所属类别的概率大于第三阈值时，也即信道鉴别模型和待训练信道生成模型达到稳定状态，完成训练，得到训练好的以参考信号为条件的信道生成模型。
这样，数据获取装置可以单独提取出训练好的信道生成模型，得到以参考信号为条件的信道生成模型，以用于生成与参考信号样本数据关联的信道信息样本数据。
在另一种可能的实现方式中，第二生成模型可以是图11所示的以信道为条件的参考信号生成模型。具体地，图17示出了以信道为条件的参考信号生成模型的训练过程。
具体地，可以先构建待训练参考信号生成模型和待训练参考信号鉴别模型。其中，待训练参考信号生成模型和待训练参考信号鉴别模型可采用全连接网络、卷积神经网络、残差网络、自注意力机制网络中的一种或多种网络结构组成。示例性的，参考图13所示，待训练信道生成模型和待训练信道鉴别模型可以由多个全连接层组成。不同的全连接层对应不同的维度，以提取不同的数据特征。
在本实现方式中，如图17所示，待训练参考信号生成模型的输入数据(即第四数据)可以为虚拟信道信息。这里，虚拟信道信息可以是已经训练好的信道生成模型生成的信道信息样本数据。可以理解的是，将虚拟信道信息作为本实现方式中待训练参考信号生成模型的输入数据，这样，待训练参考信号生成模型可以生成与虚拟信道信息具有关联关系的参考信号，从而得到对应的参考信号-信道信息样本数据对。
如图17所示，本实现方式中的待训练参考信号鉴别模型的输入可以包括虚拟参信道信息(即第四数据)、待训练参考信号生成模型生成的虚拟参考信号(即第五数据)、真实参考信号、以及真实信道信息。
可选地，在对待训练参考信号鉴别模型进行输入时，可以将虚拟信道信息(即第四数据)和虚拟参考信号(即第五数据)进行联合输入，即将虚拟信道信息和虚拟参考信号组合为一组合数据输入。示例性的，可以将图9所示的大小为M*N*P的第一参考信号样本数据，和图10所示的大小为I*J*K的第一信道信息样本数据组合为(M+I)*(N+J)*(P+K)的三维数据，并将该(M+I)*(N+J)*(P+K)的三维数据输入待训练信道鉴别模型。
同样的，在对待训练参考信号鉴别模型进行输入时，可以将真实参考信号和真实信道信息进行联合输入，即将真实参考信号和真实信道信息组成为真实组合数据进行输入。
基于此，在训练时，可以先保持待训练参考信号生成模型的模型参数不变，通过待训练参考信号鉴别模型对虚拟信道信息(即第四数据)和待训练参考信号生成模型生成的虚拟参考信号(即第五数据)组成的组合数据进行鉴别，确定该组合数据对应的类别是真实组合数据所属类别的概率(即第二鉴别结果)。进而，以降低该概率为目的，对待训练参考信号鉴别模型的模型参数进行调整，使得待训练参考信号鉴别模型尽可能地区分组合数据的真假，得到训练好的参考信号鉴别模型。
进一步，如图16所示，保持参考信号鉴别模型的模型参数不变，通过参考信号鉴别模型对虚拟 信道信息(即第四数据)和待训练信道生成模型生成的虚拟参考信号(即第五数据)组成的组合数据进行鉴别，确定当前组合数据的类别是上述真实组合数据所属类别的概率(即第二鉴别结果)。进而，可以以提高该概率为目的，调整待训练参考信号生成模型的模型参数，使得参考信号鉴别模型无法区分组合数据与上述真实组合数据的差别。
可以重复上述步骤，经过多次更新迭代后，当参考信号鉴别模型无法鉴别出组合数据到底是生成的还是真实的，即参考信号鉴别模型鉴别出组合数据应的类别为真实组合数据所属类别的概率大于第三阈值时，也即参考信号鉴别模型和待训练参考信号生成模型达到稳定状态，完成训练。如此，得到训练好的以信道为条件的参考信号生成模型。
这样，数据获取装置可以单独提取出训练好的参考信号生成模型，得到以信道为条件的参考信号生成模型，以用于生成与信道信息样本数据关联的参考信号样本数据。
除了上述利用独立的两个独立生成模型来生成具有关联关系的参考信号样本数据和信道信息样本数据，在一些实施例中，生成模型可以是单独的一个完整生成模型，即该生成模型可以同时生成具有关联关系的一对参考信号样本数据和信道信息样本数据。
具体地，在本申请实施例中，该生成模型可以没有独立的输入数据，即直接利用生成模型生成参考信号样本数据，以及信道信息样本数据。该生成模型的输入数据还可以是以下中的至少一项：噪声、随机数、信道类型指示信息、真实参考信号、真实参考信号的统计信息、真实信道信息、以及真实信道信息的统计信息。
另外，该生成模型的输出数据可以是参考信号样本数据和信道信息样本数据组成的一组数据。示例性的，该生成模型的输出数据可以是图9所示的大小为M*N*P的第一参考信号样本数据，和图10所示的大小为I*J*K的第一信道信息样本数据组合成的，大小为(M+I)*(N+J)*(P+K)的三维数据。
该模型的训练过程与上述实施例的训练过程类似，为了简洁，这里不再赘述。
综上所述，本申请实施例提供一种数据获取方法，具体地，数据获取装置可以获取生成模型生成的参考信号样本数据和信道信息样本数据；其中，该生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；另外，参考信号样本数据和信道信息样本数据用于训练信道估计模型。可以看到，本申请中的参考信号样本数据和信道信息样本数据是通过训练好的生成模型生成的，而生成模型是基于生成对抗网络构建，仅需少量的真实参考信号和真实信道信息便可实现生成模型的训练，避免了采集大量的真实参考信号及真实信道信息以及对信道的建模过程，大大降低了数据获取的难度和人工开销。
以上结合附图详细描述了本申请的优选实施方式，但是，本申请并不限于上述实施方式中的具体细节，在本申请的技术构思范围内，可以对本申请的技术方案进行多种简单变型，这些简单变型均属于本申请的保护范围。例如，在上述具体实施方式中所描述的各个具体技术特征，在不矛盾的情况下，可以通过任何合适的方式进行组合，为了避免不必要的重复，本申请对各种可能的组合方式不再另行说明。又例如，本申请的各种不同的实施方式之间也可以进行任意组合，只要其不违背本申请的思想，其同样应当视为本申请所公开的内容。又例如，在不冲突的前提下，本申请描述的各个实施例和/或各个实施例中的技术特征可以和现有技术任意的相互组合，组合之后得到的技术方案也应落入本申请的保护范围。
图18是本申请实施例提供的数据获取装置的结构组成示意图一，如图18所示，所述数据获取装置包括：
获取单元1801，配置为获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；
所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。
在一些实施例中，所述参考信号样本数据和所述信道信息样本数据具有关联关系。
在一些实施例中，所述真实参考信号是通过所述真实信道信息对应的信道接收得到。
在一些实施例中，所述真实参考信号和所述真实信道信息的数量为预设数量；所述预设数量小于或等于第一阈值。
在一些实施例中，所述生成模型包括第一生成模型和第二生成模型；
所述获取单元1801，具体配置为获取所述第一生成模型生成的第一数据；将所述第一数据输入至所述第二生成模型中，获取所述第二生成模型生成的第二数据；
其中，所述第一生成模型为参考信号生成模型，所述第二生成模型为信道生成模型；或者，所述第一生成模型为信道生成模型，所述第二生成模型为参考信号生成模型；所述参考信号生成模型 用于生成所述参考信号样本数据，所述信道生成模型用于生成所述信道信息样本数据。
在一些实施例中，还包括第一训练单元，配置为：获取待训练第一生成模型输出的第三数据；将所述第三数据输入第一鉴别模型，通过所述第一鉴别模型输出第一鉴别结果；所述第一鉴别模型用于鉴别所述第三数据对应的类别为第一真实数据所属类别的概率；所述第一真实数据与所述第一生成模型具有关联关系；基于所述第一鉴别结果，对所述待训练第一生成模型的模型参数进行调整，得到所述第一生成模型；所述第一生成模型生成的数据对应的类别为所述第一真实数据所属类别的概率大于第二阈值。
在一些实施例中，所述第一生成模型为参考信号生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
噪声、随机数、信道类型指示信息、真实参考信号、以及真实参考信号的统计信息。
在一些实施例中，所述第一生成模型为信道信息生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
噪声、随机数、信道类型指示信息、真实信道信息、以及真实信道信息的统计信息。
在一些实施例中，所述信道类型指示信息包括以下中的至少一项：
第一指示信息，用于指示信道所处的频段；
第二指示信息，用于指示射频环境；
第三指示信息，用于指示无线信道场景。
在一些实施例中，还包括第二训练单元，配置为将第四数据输入至待训练第二生成模型中，通过所述待训练第二生成模型生成第五数据，所述第四数据为所述第一生成模型生成的数据；将所述第五数据输入第二鉴别模型，通过所述第二鉴别模型输出第二鉴别结果；所述第二鉴别结果用于鉴别组合数据对应的类别为第二真实数据所属类别的概率；所述组合数据包括所述第四数据和所述第五数据；所述第二真实数据包括所述真实参考信号和所述真实信道信息；基于所述第二鉴别结果，对所述待训练第二生成模型的模型参数进行调整，得到所述第二生成模型；所述第二生成模型生成的数据与所述第四数据组合后所形成的数据，其对应的类别与所述第二真实数据所属类别的概率大于第三阈值。
在一些实施例中，所述第二生成模型为信道信息生成模型，所述第四数据为参考信号生成模型生成的数据。
在一些实施例中，所述第二生成模型为参考信号生成模型，所述第四数据为信道信息生成模型生成的数据。
本领域技术人员应当理解，本申请实施例的上述数据获取装置的相关描述可以参照本申请实施例的数据获取方法的相关描述进行理解。
图19是本申请实施例提供的一种电子设备1900示意性结构图。该电子设备可以是三方的服务器、也可以是终端设备，或网络设备，本申请实施例对此不做限定。图19所示的电子设备1900包括处理器1910，处理器1910可以从存储器中调用并运行计算机程序，以实现本申请实施例中的方法。
可选地，如图19所示，电子设备1900还可以包括存储器1920。其中，处理器1910可以从存储器1920中调用并运行计算机程序，以实现本申请实施例中的数据获取方法。
其中，存储器1920可以是独立于处理器1910的一个单独的器件，也可以集成在处理器1910中。
图20是本申请实施例的芯片的示意性结构图。图20所示的芯片2000包括处理器2010，处理器2010可以从存储器中调用并运行计算机程序，以实现本申请实施例中的方法。
可选地，如图20所示，芯片2000还可以包括存储器2020。其中，处理器2010可以从存储器2020中调用并运行计算机程序，以实现本申请实施例中的方法。
其中，存储器2020可以是独立于处理器2010的一个单独的器件，也可以集成在处理器2010中。
可选地，该芯片2000还可以包括输入接口2030。其中，处理器2010可以控制该输入接口2030与其他设备或芯片进行通信，具体地，可以获取其他设备或芯片发送的信息或数据。
可选地，该芯片2000还可以包括输出接口2040。其中，处理器2010可以控制该输出接口2040与其他设备或芯片进行通信，具体地，可以向其他设备或芯片输出信息或数据。
可选地，该芯片可应用于本申请实施例中的网络设备，并且该芯片可以实现本申请实施例的各个方法中由网络设备实现的相应流程，为了简洁，在此不再赘述。
可选地，该芯片可应用于本申请实施例中的移动终端/终端设备，并且该芯片可以实现本申请实施例的各个方法中由移动终端/终端设备实现的相应流程，为了简洁，在此不再赘述。
应理解，本申请实施例提到的芯片还可以称为系统级芯片，系统芯片，芯片系统或片上系统芯片等。
应理解，本申请实施例的处理器可能是一种集成电路芯片，具有信号的处理能力。在实现过程中，上述方法实施例的各步骤可以通过处理器中的硬件的集成逻辑电路或者软件形式的指令完成。上述的处理器可以是通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。结合本申请实施例所公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器，处理器读取存储器中的信息，结合其硬件完成上述方法的步骤。
可以理解，本申请实施例中的存储器可以是易失性存储器或非易失性存储器，或可包括易失性和非易失性存储器两者。其中，非易失性存储器可以是只读存储器(Read-Only Memory，ROM)、可编程只读存储器(Programmable ROM，PROM)、可擦除可编程只读存储器(Erasable PROM，EPROM)、电可擦除可编程只读存储器(Electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取存储器(Random Access Memory，RAM)，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(Static RAM，SRAM)、动态随机存取存储器(Dynamic RAM，DRAM)、同步动态随机存取存储器(Synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM，DDR SDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(Synchlink DRAM，SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM，DR RAM)。应注意，本文描述的系统和方法的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
应理解，上述存储器为示例性但不是限制性说明，例如，本申请实施例中的存储器还可以是静态随机存取存储器(static RAM，SRAM)、动态随机存取存储器(dynamic RAM，DRAM)、同步动态随机存取存储器(synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM，DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(synch link DRAM，SLDRAM)以及直接内存总线随机存取存储器(Direct Rambus RAM，DR RAM)等等。也就是说，本申请实施例中的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
本申请实施例还提供了一种计算机可读存储介质，用于存储计算机程序。
可选的，该计算机可读存储介质可应用于本申请实施例中的网络设备，并且该计算机程序使得计算机执行本申请实施例的各个方法中由电子设备实现的相应流程，为了简洁，在此不再赘述。
本申请实施例还提供了一种计算机程序产品，包括计算机程序指令。
可选的，该计算机程序产品可应用于本申请实施例中的电子设备，并且该计算机程序指令使得计算机执行本申请实施例的各个方法中由网络设备实现的相应流程，为了简洁，在此不再赘述。
本申请实施例还提供了一种计算机程序。
可选的，该计算机程序可应用于本申请实施例中的电子设备，当该计算机程序在计算机上运行时，使得计算机执行本申请实施例的各个方法中由网络设备实现的相应流程，为了简洁，在此不再赘述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它 的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，)ROM、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应所述以权利要求的保护范围为准。

Claims (29)

  1. 一种数据获取方法，所述方法包括：
    获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；
    所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。
  2. 根据权利要求1所述的方法，其中，所述参考信号样本数据和所述信道信息样本数据具有关联关系。
  3. 根据权利要求1或2所述的方法，其中，所述真实参考信号是通过所述真实信道信息对应的信道接收得到。
  4. 根据权利要求1-3任一项所述的方法，其中，所述真实参考信号和所述真实信道信息的数量为预设数量；所述预设数量小于或等于第一阈值。
  5. 根据权利要求1-4任一项所述的方法，其中，所述生成模型包括第一生成模型和第二生成模型，所述获取生成模型生成的参考信号样本数据和信道信息样本数据，包括：
    获取所述第一生成模型生成的第一数据；
    将所述第一数据输入至所述第二生成模型中，获取所述第二生成模型生成的第二数据；
    其中，所述第一生成模型为参考信号生成模型，所述第二生成模型为信道生成模型；或者，所述第一生成模型为信道生成模型，所述第二生成模型为参考信号生成模型；所述参考信号生成模型用于生成所述参考信号样本数据，所述信道生成模型用于生成所述信道信息样本数据。
  6. 根据权利要求5所述的方法，其中，所述第一生成模型的训练过程包括：
    获取待训练第一生成模型输出的第三数据；
    将所述第三数据输入第一鉴别模型，通过所述第一鉴别模型输出第一鉴别结果；所述第一鉴别模型用于鉴别所述第三数据对应的类别为第一真实数据所属类别的概率；所述第一真实数据与所述第一生成模型具有关联关系；
    基于所述第一鉴别结果，对所述待训练第一生成模型的模型参数进行调整，得到所述第一生成模型；所述第一生成模型生成的数据对应的类别为所述第一真实数据所属类别的概率大于第二阈值。
  7. 根据权利要求5或6所述的方法，其中，所述第一生成模型为参考信号生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
    噪声、随机数、信道类型指示信息、真实参考信号、以及真实参考信号的统计信息。
  8. 根据权利要求5或6所述的方法，其中，所述第一生成模型为信道信息生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
    噪声、随机数、信道类型指示信息、真实信道信息、以及真实信道信息的统计信息。
  9. 根据权利要求7或8所述的方法，其中，所述信道类型指示信息包括以下中的至少一项：
    第一指示信息，用于指示信道所处的频段；
    第二指示信息，用于指示射频环境；
    第三指示信息，用于指示无线信道场景。
  10. 根据权利要求5所述的方法，其中，所述第二生成模型的训练过程包括：
    将第四数据输入至待训练第二生成模型中，通过所述待训练第二生成模型生成第五数据，所述第四数据为所述第一生成模型生成的数据；
    将所述第五数据输入第二鉴别模型，通过所述第二鉴别模型输出第二鉴别结果；所述第二鉴别结果用于鉴别组合数据对应的类别为第二真实数据所属类别的概率；所述组合数据包括所述第四数据和所述第五数据；所述第二真实数据包括所述真实参考信号和所述真实信道信息；
    基于所述第二鉴别结果，对所述待训练第二生成模型的模型参数进行调整，得到所述第二生成模型；所述第二生成模型生成的数据与所述第四数据组合后所形成的数据，其对应的类别与所述第二真实数据所属类别的概率大于第三阈值。
  11. 根据权利要求10所述的方法，其中，所述第二生成模型为信道信息生成模型，所述第四数据为参考信号生成模型生成的数据。
  12. 根据权利要求10所述的方法，其中，所述第二生成模型为参考信号生成模型，所述第四数据为信道信息生成模型生成的数据。
  13. 一种数据获取装置，包括：
    获取单元，配置为获取生成模型生成的参考信号样本数据和信道信息样本数据；所述生成模型基于生成对抗网络、真实参考信号、以及真实信道信息训练得到；
    所述参考信号样本数据和信道信息样本数据用于训练信道估计模型。
  14. 根据权利要求13所述的装置，其中，所述参考信号样本数据和所述信道信息样本数据具有关联关系。
  15. 根据权利要求13或14所述的装置，其中，所述真实参考信号是通过所述真实信道信息对应的信道接收得到。
  16. 根据权利要求13-15任一项所述的装置，其中，所述真实参考信号和所述真实信道信息的数量为预设数量；所述预设数量小于或等于第一阈值。
  17. 根据权利要求13-16任一项所述的装置，其中，所述生成模型包括第一生成模型和第二生成模型；
    所述获取单元，具体配置为获取所述第一生成模型生成的第一数据；将所述第一数据输入至所述第二生成模型中，获取所述第二生成模型生成的第二数据；
    其中，所述第一生成模型为参考信号生成模型，所述第二生成模型为信道生成模型；或者，所述第一生成模型为信道生成模型，所述第二生成模型为参考信号生成模型；所述参考信号生成模型用于生成所述参考信号样本数据，所述信道生成模型用于生成所述信道信息样本数据。
  18. 根据权利要求17所述的装置，其中，还包括第一训练单元，配置为：获取待训练第一生成模型输出的第三数据；将所述第三数据输入第一鉴别模型，通过所述第一鉴别模型输出第一鉴别结果；所述第一鉴别模型用于鉴别所述第三数据对应的类别为第一真实数据所属类别的概率；所述第一真实数据与所述第一生成模型具有关联关系；基于所述第一鉴别结果，对所述待训练第一生成模型的模型参数进行调整，得到所述第一生成模型；所述第一生成模型生成的数据对应的类别为所述第一真实数据所属类别的概率大于第二阈值。
  19. 根据权利要求17或18所述的装置，其中，所述第一生成模型为参考信号生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
    噪声、随机数、信道类型指示信息、真实参考信号、以及真实参考信号的统计信息。
  20. 根据权利要求17或18所述的装置，其中，所述第一生成模型为信道信息生成模型，所述第一生成模型在生成数据的过程中无输入数据，或者所述第一生成模型以以下中的至少一项数据作为输入数据：
    噪声、随机数、信道类型指示信息、真实信道信息、以及真实信道信息的统计信息。
  21. 根据权利要求19或20所述的装置，其中，所述信道类型指示信息包括以下中的至少一项：
    第一指示信息，用于指示信道所处的频段；
    第二指示信息，用于指示射频环境；
    第三指示信息，用于指示无线信道场景。
  22. 根据权利要求17所述的装置，其中，还包括第二训练单元，配置为将第四数据输入至待训练第二生成模型中，通过所述待训练第二生成模型生成第五数据，所述第四数据为所述第一生成模型生成的数据；将所述第五数据输入第二鉴别模型，通过所述第二鉴别模型输出第二鉴别结果；所述第二鉴别结果用于鉴别组合数据对应的类别为第二真实数据所属类别的概率；所述组合数据包括所述第四数据和所述第五数据；所述第二真实数据包括所述真实参考信号和所述真实信道信息；基于所述第二鉴别结果，对所述待训练第二生成模型的模型参数进行调整，得到所述第二生成模型；所述第二生成模型生成的数据与所述第四数据组合后所形成的数据，其对应的类别与所述第二真实数据所属类别的概率大于第三阈值。
  23. 根据权利要求22所述的装置，其中，所述第二生成模型为信道信息生成模型，所述第四数据为参考信号生成模型生成的数据。
  24. 根据权利要求22所述的装置，其中，所述第二生成模型为参考信号生成模型，所述第四数据为信道信息生成模型生成的数据。
  25. 一种电子设备，包括：存储器和处理器，
    所述存储器存储有可在处理器上运行的计算机程序，
    所述处理器执行所述程序时实现权利要求1至12任一项所述方法。
  26. 一种计算机存储介质，所述计算机存储介质存储有一个或者多个程序，所述一个或者多个程序可被一个或者多个处理器执行，以实现权利要求1至12任一项所述方法。
  27. 一种芯片，包括：处理器，用于从存储器中调用并运行计算机程序，使得安装有所述芯片的设备执行如权利要求1至12任一项所述方法。
  28. 一种计算机程序产品，所述计算机程序产品包括计算机存储介质，所述计算机存储介质存储计算机程序，所述计算机程序包括能够由至少一个处理器执行的指令，当所述指令由所述至少一个处理器执行时实现权利要求1至12任一项所述方法。
  29. 一种计算机程序，所述计算机程序使得计算机执行如权利要求1至12任一项所述方法。
PCT/CN2021/135300 2021-12-03 2021-12-03 数据获取方法、装置、设备、介质、芯片、产品及程序 WO2023097645A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/135300 WO2023097645A1 (zh) 2021-12-03 2021-12-03 数据获取方法、装置、设备、介质、芯片、产品及程序

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/135300 WO2023097645A1 (zh) 2021-12-03 2021-12-03 数据获取方法、装置、设备、介质、芯片、产品及程序

Publications (1)

Publication Number Publication Date
WO2023097645A1 true WO2023097645A1 (zh) 2023-06-08

Family

ID=86611330

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/135300 WO2023097645A1 (zh) 2021-12-03 2021-12-03 数据获取方法、装置、设备、介质、芯片、产品及程序

Country Status (1)

Country Link
WO (1) WO2023097645A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016106753A1 (zh) * 2014-12-31 2016-07-07 华为技术有限公司 一种无线局域网的信道估计方法和装置
WO2017155634A1 (en) * 2016-03-11 2017-09-14 Origin Wireless, Inc. Methods, devices, servers, apparatus, and systems for wireless internet of things applications
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法
CN112787966A (zh) * 2020-12-28 2021-05-11 杭州电子科技大学 基于端到端的级联生成对抗网络信号解调方法
CN113381952A (zh) * 2021-06-09 2021-09-10 东南大学 基于深度学习的多天线系统信道估计方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016106753A1 (zh) * 2014-12-31 2016-07-07 华为技术有限公司 一种无线局域网的信道估计方法和装置
WO2017155634A1 (en) * 2016-03-11 2017-09-14 Origin Wireless, Inc. Methods, devices, servers, apparatus, and systems for wireless internet of things applications
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法
CN112787966A (zh) * 2020-12-28 2021-05-11 杭州电子科技大学 基于端到端的级联生成对抗网络信号解调方法
CN113381952A (zh) * 2021-06-09 2021-09-10 东南大学 基于深度学习的多天线系统信道估计方法

Similar Documents

Publication Publication Date Title
EP4262121A1 (en) Neural network training method and related apparatus
WO2021001032A1 (en) Transmission system with channel estimation based on a neural network
Safari et al. Deep UL2DL: Data-driven channel knowledge transfer from uplink to downlink
WO2016150246A1 (zh) 一种信号检测方法及装置
CN107864029A (zh) 一种降低多用户检测复杂度的方法
Wimalajeewa et al. Distributed asynchronous modulation classification based on hybrid maximum likelihood approach
WO2022094778A1 (zh) 多输入多输出mimo系统的检测方法和装置
CN108206795B (zh) 基于置信传递的频率选择性衰落信道的盲认证方法和系统
CN117813913A (zh) 使用神经网络进行信源编码的方法和系统
US11929853B2 (en) Data-driven probabilistic modeling of wireless channels using conditional variational auto-encoders
Struminsky et al. A new approach for sparse Bayesian channel estimation in SCMA uplink systems
Drakshayini et al. A review of wireless channel estimation techniques: challenges and solutions
CN108566227B (zh) 一种多用户检测方法
US20240048413A1 (en) Channel restoration method and receiving device
WO2008151518A1 (fr) Procédé et dispositif de détection d'information dans un système ofdm
WO2023097645A1 (zh) 数据获取方法、装置、设备、介质、芯片、产品及程序
CN108173791A (zh) 基于平滑技术的时变衰落信道的物理层盲认证方法及系统
WO2023125660A1 (zh) 一种通信方法及装置
WO2022206328A1 (zh) 一种通信协作方法及装置
Bhatt et al. Analysis of the fifth generation NOMA system using LSTM algorithm
WO2022001822A1 (zh) 获取神经网络的方法和装置
CN108199991A (zh) 基于置信传递的时变衰落信道的物理层盲认证方法和系统
WO2023060503A1 (zh) 信息处理方法、装置、设备、介质、芯片、产品及程序
WO2023092307A1 (zh) 通信方法、模型训练方法和设备
WO2023070675A1 (zh) 数据处理的方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21966072

Country of ref document: EP

Kind code of ref document: A1